Daily Digest

News

Daily headlines drawn from across every company we track: the last 7 days, with every earlier story in the archive.

/ Latest Company Overviews

/ Saturday, November 21, 2026

Yicai says AgiBot put G2 robots on a phone testing line

Yicai reported on Nov. 20 that its reporter had learned AgiBot and Lens Technology recently agreed to deploy G2 robots, called 精灵G2 in the brief, on a phone testing line, with dozens of units loading and unloading stations and interacting with MMI, ANT, CAM and CAMCV equipment. The brief also restated an earlier Longcheer Nanchang deployment. Neither company had published the agreement in the pages opened, and the brief gives no site, contract value or operating hours.

Agility names a president and two operating chiefs

Agility said on Nov. 18 that it promoted Daniel Diez from chief business officer to president, appointed Christopher Sullivan chief product officer, and named Ana Lang chief administrative officer. The release quotes Peggy Johnson as chief executive. Diez will lead the commercial business, Sullivan will oversee product and customer success, and Lang will keep legal and people duties as general counsel and corporate secretary.

An Anthropic mathematician says Claude helped prove the Riemann hypothesis

Levent Alpöge wrote on Nov. 20 that he and Claude had found a proof of the Riemann hypothesis and that the method also resolves the generalized hypothesis, and he posted the argument as a Lean formalization in 14 parts. Nature reported that Anthropic did not immediately comment and that Andrew Granville called the four-page sketch garbled. Andrew Sutherland at MIT reported on Nov. 21 that the Lean source compiles and that separate kernels accept five headline theorems, covering the Riemann hypothesis and Dirichlet L-functions, under only the three standard axioms; some kernel replays were still running, other L-functions were outside the check, and the Clay prize process had not accepted the argument.

Falcon 9 launches the classified USSF-317 mission

SpaceX said Falcon 9 launched the classified USSF-317 mission from Vandenberg Space Launch Complex 4E at 8:20 p.m. PT on Nov. 20, and that the first stage landed on the Of Course I Still Love You droneship on its 11th flight. Space.com designates the booster B1100 and reports the same liftoff and landing. SpaceX's page does not state payload deployment or orbit, and this account does not use a satellite count.

Oracle says its OpenAI infrastructure contract is funded

Oracle said on Nov. 20 that it has the funding in place to fulfill its infrastructure contract with OpenAI, including the vendor financing described at Investor Day, and that further vendor talks are optional. Oracle said the Abilene campus, where Astra-6 was trained, is nearly complete, that its next gigawatt-scale campus at Shackelford is fully permitted, and that Nvidia Vera Rubin racks there are on track for handover in the coming weeks. OpenAI has not confirmed the statement. The Oracle page says Astra-6; Reuters describes the trained model as GPT-6 Astra.

Huawei launches the OSTA O1 Pro from 5,499 yuan

Huawei launched the OSTA phone brand on Nov. 20 with the O1 Pro, Caixin and TechNode reported. Caixin said prices run from 5,499 yuan to 8,799 yuan through Dec. 31 and that the phone runs HarmonyOS. TechNode, citing STAR Market Daily, said the wide-screen model uses a Kirin 9030S and an AI-focused edition of HarmonyOS 7. This pass did not open a Huawei product page.

Google will retire the Nest app on Jan. 25, 2027

Google said the Nest app on Android, iOS, TV, and home.nest.com will retire on Jan. 25, 2027, with control moving to the Google Home app and home.google.com. Supported Nest devices and camera video history are set to transfer automatically, saved web clips will not, and Nest Aware first-generation plans are set to move to Google Home Premium Advanced at no added cost. Google also said Assistant on speakers and smart displays will retire on Aug. 2, 2027, replaced by Gemini for Home, and that a Nest account must be migrated to a Google Account before data deletions begin on March 25, 2027.

Microsoft will retire To Do apps on June 30, 2027

Microsoft said To Do apps on the web, Windows, Mac, iOS, and Android will retire on June 30, 2027, for work, school, and personal accounts. Work and school users are directed to Planner, where an updated My Tasks experience begins rolling out in December 2026, but existing To Do tasks will not migrate and stay viewable in Planner only through that date. Planner is not available for personal accounts, and Microsoft said the export instructions and the later API retirement dates are not yet published.

WeRide reports Q3 revenue of RMB 321.5 million

WeRide reported unaudited third-quarter revenue of RMB 321.5 million (US$48.0 million), up 88.0 percent year over year, with a 43.0 percent gross margin and a loss of RMB 310.0 million. Robotaxi revenue exceeded RMB 100 million, and the company said its global L4 fleet was about 4,100 vehicles as of Oct. 31, including more than 2,300 robotaxis. Cumulative deliveries of its WRD 3.0 driver-assistance system passed 100,000 units by the end of the quarter.

New Mexico licensed Pacific Fusion to hold tritium

Pacific Fusion said on Nov. 20 that the New Mexico Environment Department approved a radioactive-materials license for its Albuquerque campus, which the company called the first the state has issued to a private fusion company. The Albuquerque Journal reported that the license was granted in August, allows up to 15,000 curies of tritium (about 1.5 grams), and that the company does not plan to hold more than 10,000 curies at a time. The Mesa del Sol campus is still under construction, and the company said the tritium would stay in a closed system.

Tesla seeks a tax limit for a $3.1 billion Texas polysilicon plant

Tesla has applied for a Texas school-tax limit on a proposed $3.1 billion polysilicon refinery near Robstown, Nueces County, according to a Jobs, Energy, Technology and Innovation filing described by pv magazine and KRIS 6. The company calls the plant under consideration, says it would not invest in Texas without the agreement, and is still comparing other sites. An EY analysis in the filing schedules construction from 2027 to 2030 and 515 jobs by 2032, with no production capacity stated, and the school board and governor have not approved a deal.

Long March-8A orbits the 30th low-orbit internet group

A Long March-8A lifted off at 7:59 p.m. Beijing time on Nov. 20 from the Hainan commercial launch site and placed the 30th group of low-orbit internet satellites into the designated orbit, CASC's China Space News report and Xinhua said. CASC numbered the flight the 679th Long March mission. Neither release states how many satellites were aboard or their altitude.

Oklo says a Lebanon isotope plan would not make electricity

Oklo told the Indiana Capital Chronicle it is still evaluating a potential isotope facility near Lebanon and has not decided to build there. The company said it does not plan an electricity-producing reactor at the LEAP site and confirmed it is cutting the possible land purchase from 440 acres to 220. The state development agency said the revised agreement is still being negotiated, and a potential investment of up to $1.5 billion remains uncommitted.

/ Friday, November 20, 2026

CRS-35 Dragon undocks from the space station

NASA said the unpiloted SpaceX CRS-35 Dragon undocked from the Harmony module at 12:05 p.m. EST on Nov. 20, 2026, after a command from SpaceX controllers. The station blog targets a splashdown off California at about 4:55 p.m. EST on Nov. 21 and says NASA will not stream it. Space.com reported the same undock time and said the departure webcast missed the moment of separation.

OpenAI publishes the MedChemBench chemistry benchmark

On Nov. 20, 2026 OpenAI introduced MedChemBench, 99 problems drawn from 117 papers and patents and 761 molecular structures, covering structure reconstruction, reasoning, compound selection, and property prediction. At the xhigh setting, OpenAI says GPT-6.2 Astra scored 84.5 percent of available credit, against 83.3 percent for GPT-6 Astra, 82.0 percent for GPT-6.1 Sol, and 58.2 percent for GPT-5.6 Sol. OpenAI says the scores do not show the models can discover drugs, and that prior exposure to the published sources cannot be ruled out.

SK hynix plans to convert Kioxia bonds into a 14.1 percent stake

SK hynix said in a Nov. 20, 2026 filing that it will convert 129 billion yen of bonds in BCPE Pangea Cayman2 into 129,000 common shares and buy 1,001 further shares from Bain Capital for 100 million yen. On completion, Yonhap reported, the company would hold an indirect 14.1 percent of Kioxia, ahead of Toshiba's 10.46 percent as of early November, without plans to buy or sell more shares or to expand its management role. Channel NewsAsia placed the step in December, and the Korea Herald reported a Dec. 4 conversion date.

Reuters says the SEC is examining Nvidia's accounting

Reuters reported on Nov. 20, 2026 that two people said SEC enforcement staff are examining potential accounting issues at Nvidia, after the Wall Street Journal had reported an SEC probe without identifying its focus. The SEC declined to comment, Nvidia had not responded, and Reuters said it could not determine when the inquiry opened or which practices are under review. The report is not a finding, and it is separate from California's Nov. 19 investigative subpoena.

Samsung names new heads of memory and mobile

Samsung said on Nov. 20, 2026 that it promoted Sang Joon Hwang to president and head of Memory, Sangyeun Cho to president and head of System LSI, and Jong Myeong Lee to president and chief technology officer of Device Solutions, and named Won-Joon Choi head of Mobile eXperience. Young Hyun Jun and TM Roh remain chief executives of the Device Solutions and Device eXperience divisions. The Korea Herald reported that the two are leaving the memory and mobile unit posts, with Jun focusing on foundry and Roh on new businesses including robotics.

Italy provisionally approves Tesla Assisted Driving

Italy's transport minister, Matteo Salvini, said on Nov. 20, 2026 that the ministry provisionally approved Tesla Assisted Driving after a technical feasibility assessment, ahead of an EU recognition process that is still being defined. Reuters called Italy the first major EU country to take that step and said Tesla posted that the system was approved, with a video from Rome. Electrek reported a Tesla Europe line that rollout begins soon, with no date, and that the driver must still supervise the car.

Arrangers price about $44 billion of notes for SpaceX chips

Apollo and Goldman Sachs said on Nov. 20, 2026 that they priced about $44 billion of senior and subordinated notes, oversubscribed and upsized, to buy Nvidia Vera Rubin chips and related equipment for SpaceX's terrestrial compute, with lease schedules expected to start on delivery in the first half of 2027. IFR reported about $32.1 billion of public bonds, including $25.168 billion due April 2033 at a yield of 8.259 percent, plus a senior loan and private notes. SpaceX has not posted the pricing, and its Oct. 24 letter still put principal at about $40 billion on a lease that had not been signed.

Amazon Polly offers gated voice cloning in seven languages

On Nov. 19, 2026 AWS said voice cloning in Amazon Polly is generally available as a gated release. A customer sends 1 to 10 recordings, up to 15 minutes in total, of one consenting speaker; AWS said creation usually takes 5 to 10 minutes, the synthesized audio carries a watermark, and AWS deletes its copy of the recordings after creation and does not use them to train Polly models. Each cloned voice speaks the language it was recorded in. The launch covers English, Spanish, German, French, Italian, Portuguese, and Hindi. Building a voice is free and synthesis is pay as you go. New accounts must request access, and the initial quota is 10 cloned voices per region.

California serves Nvidia an investigative subpoena

On Nov. 19, 2026 California Attorney General Rob Bonta said his office served Nvidia with an investigative subpoena. The release ties it to news reports of an affair between former CEO Jensen Huang and an alleged Chinese spy. The office is asking whether advanced chips or confidential information may have gone to China and whether Nvidia misled investors and the public. It says the answers will help evaluate whether California law was violated, and it does not state a finding.

Anthropic closes its IPO and the extra-share option

On Nov. 20, 2026 Anthropic said its Class A offering closed at $83.00 a share, including the underwriters' full option to buy 147,000,000 more shares from the company. The closed offering is 1,127,000,000 shares: 1,103,351,354 sold by Anthropic and 23,648,646 by selling stockholders. Anthropic said gross proceeds before discounts and expenses were about $93.5 billion, about $91.6 billion of that for the company, and that the $26.0 billion Class N sale to AMD, Broadcom, Google, Micron, Microsoft, Nvidia, and TSMC Partners also closed. Reuters reported the option exercise and the roughly $93.5 billion total. Micron is not in the catalog.

A UK tribunal says the Home Office gave Apple a capability notice

On Nov. 19, 2026 the Investigatory Powers Tribunal held that the Home Secretary had not shown it would damage national security to confirm, in these proceedings, that she has given Apple a technical capability notice under the Investigatory Powers Act 2016. The open judgment says that confirmation can be made in public unless a separate secrecy basis applies. It says the notice's precise content has not been published, and it does not decide whether the notice is lawful. Apple's challenge remains before the tribunal.

OpenAI says coding agents finish 46 percent of mid-length research tasks

On Nov. 18, 2026 OpenAI published an internal update on research acceleration. It said estimated success without human intervention rose from 23 percent in March to 46 percent in October for tasks it estimates take a person two to four hours, and from 13 percent to 35 percent for four-to-eight-hour tasks. For Aug. 1 through Oct. 18 it put success at 79 percent for monitoring training runs, 35 percent for launching them, and 25 percent for inference and serving code. OpenAI said the rates can also reflect which tasks researchers delegate, and that the report is not a plan to pursue full recursive self-improvement.

Google previews Tool Retrieval in the Gemini API

On Nov. 19, 2026 Google launched Tool Retrieval in preview in the Gemini Interactions API. Deferred tools keep their names and descriptions in context, and the API loads a parameter schema only when the model asks for that tool, inside the same request. Google says there is no separate charge and that the feature works with Gemini 3.6 and later models, including every Gemini 4 model.

Fang Cheng Bao lists the Tai 9 from 279,900 yuan

Autohome and ZOL report that Fang Cheng Bao listed the Tai 9 on Nov. 19, 2026, in three versions at 279,900, 299,900, and 329,900 yuan. Autohome calls it the first model on BYD's DMS hybrid platform and quotes a 66.48 kWh battery, 310 km of WLTC electric range, and dual motors totaling 400 kW. ZOL also quotes 400 kW, a 400 km CLTC electric range, and a combined range of 1,630 km.

A short seller alleges AMD firmware reached a blacklisted server maker

On Nov. 19, 2026 Hunterbrook, which disclosed a short position in AMD, said IEIT Systems had incorporated at least six newer AMD firmware versions after the Commerce Department added IEIT to the Entity List in March 2025, most recently in a post dated Oct. 19. Hunterbrook said it does not know how IEIT obtained the code or whether anyone held a license. AMD later told CNBC the claims misrepresent its practices and that it complies with U.S. export rules. CNBC said it did not independently verify the report and that the shares closed down 7.72 percent at $489.19.

OpenAI adds GPT-Live-1.1 to the voice API

On Nov. 19, 2026 OpenAI's API changelog listed GPT-Live-1.1 for Live sessions. The model page says it can listen while it speaks, transcribes more accurately than GPT-Live-1, and keeps a more consistent accent across languages. Active voice time is $0.05 a minute, and WebRTC standby, which pauses inference during silence, is $0.02 a minute. Backend model and tool use is billed separately.

A China Telecom lab reports Xing 4 training on Ascend chips

China Telecom's Xingchen AGI Lab said in a Nov. 17 preprint that it pretrained a 285-billion-parameter Xing 4 model, with 17 billion parameters active per token, on 8,192 Ascend 910C cards. The lab said the work was done with Huawei and Harbin Institute of Technology, Shenzhen, and that it open-sourced a 29-billion-parameter model trained on Ascend with MindSpore. Its 4.68-times prefill comparison is a simulation against an MLA-only baseline, and the authors say simulation can be more optimistic than full deployment.

Ford ships the first cells from its CATL-licensed Michigan line

Ford said on Nov. 18 that the first lithium-iron-phosphate prismatic cells from BlueOval Battery Park Michigan are going to Louisville for Fathom pickup prototypes. Ford owns and operates the plant and said it uses technology licensed from CATL, with lines that can make 40 cells a minute. CNBC reported a $3 billion plant, production that started in October, expected capacity of 20 gigawatt-hours a year after a cut from up to 40, and a Fathom launch targeted for the fourth quarter of 2027.

Microsoft makes Azure Cobalt 200 virtual machines generally available

Microsoft said at Ignite on Nov. 17 that Azure Cobalt 200 virtual machines are generally available. The Azure post says they are in 14 regions, deliver up to 50 percent better performance than Cobalt 100, and use a TSMC 3-nanometer chip with 132 Arm Neoverse V3 cores. That post says availability will reach 23 regions by the end of 2026, while the Ignite book of news says 24. Microsoft quoted an OpenAI engineer saying the same ChatGPT workloads showed higher throughput and lower latency on Cobalt 200 in a test.

Varda recovers the W-9 capsule at Koonibba

Varda said W-9 landed at the Koonibba Test Range at 5:56 a.m. Pacific on Nov. 17 after 48 days, its second reentry this month, carrying a government-funded hypersonic navigation system and heat-shield samples. SatNews, quoting Southern Launch, puts the recovery at 12:06 a.m. ACDT on Nov. 18, the date the Australian Space Agency records, while Southern Launch's mission page still lists an open window. W-7 was still in orbit, with a return planned before the end of the year.

Google adds a Robotics plugin to Antigravity

Google said on Nov. 17 that a Robotics plugin for Antigravity gives Gemini 4 Hydrogen skills for robot bring-up, MuJoCo simulation, and closed-loop control. The company says Hydrogen scored 58 on the Artificial Analysis Robotics 0.1 benchmark while controlling a real six-axis arm through eight tabletop tasks. Antigravity requires a paid Google AI Pro or Ultra subscription, and Google stated no separate plugin price.

Tesla opens Cybercab rides in Dallas

Tesla's Robotaxi account said on Nov. 19 that Cybercab had arrived in Dallas, and Teslarati and InsideEVs reported that riders can book the two-seat vehicle, which has no steering wheel or pedals, in the Robotaxi app. Dallas is the second Cybercab city after Austin. Tesla did not state a Dallas fleet size or fare.

Gemini on Mac can operate local apps and the web

Google said on Nov. 19 that the Gemini app for Mac can complete multi-step tasks across local files, desktop apps, and the web inside an isolated background window. Users choose which apps it can access, can pause it, and approve key actions. It is rolling out in English in the United States for personal Google AI Plus, Pro, and Ultra accounts on Apple silicon Macs running macOS Sequoia 15 or later.

TSMC is reported to halt some advanced China orders

Reuters reported on Nov. 19, citing two people and TSMC guidance they reviewed, that affected orders at 16 nanometres and below are to stop unless customers obtain U.S. export licenses, with those Chinese designers generally limited to 28- and 40-nanometre processes if the chips do not use high-bandwidth memory or advanced packaging. A Central News Agency account of a Financial Times report said Samsung, as well as TSMC, told Chinese customers that some processor shipments would pause during license applications. Samsung declined to comment, TSMC did not respond, and neither company posted its own confirmation.

/ Thursday, November 19, 2026

Meta pauses Class C shares during a shareholder case

Meta agreed in a Delaware filing on Nov. 18, 2026 not to issue non-voting Class C shares, by dividend or otherwise, while shareholder lawsuits are pending. The filing still allows the charter that creates the class to take effect on Jan. 19, 2027, and it proposes a trial for April 19-23, 2027, subject to the chancellor's approval. Bloomberg reported that the pause covers distribution of the class Meta had planned to use in funding AI spending.

Alibaba reports RMB 4.4 billion for the Qwen segment

On Nov. 19, 2026 Alibaba said AI Labs and Applications revenue was RMB 4,435 million, which it also stated as US$661 million, for the quarter ended Sept. 30, up 51 percent from a year earlier. The results say that segment holds the AI model labs, the Qwen consumer business and QwenWork. Adjusted EBITA for the segment was a loss of RMB 10,503 million, wider than the year-earlier loss of RMB 3,435 million.

Nvidia says it is investigating a partner incident

On Nov. 19, 2026 Nvidia said it is investigating an incident that may have affected a third-party partner and that it has found no indication its own systems or networks were affected. The statement followed a listing by the Gentlemen ransomware group, which had not posted proof of stolen Nvidia data. Nvidia did not name the partner or say whether any Nvidia data was exposed.

BETA signs Astronics for production power systems

On Nov. 17, 2026 Astronics said it signed a production agreement to supply CorePower electrical power systems for BETA's ALIA CX300, ALIA A250 and HV250 aircraft. BETA's chief operating officer said the agreement moves those flight-critical systems into production. The release does not state a price, a quantity or a delivery date, and Astronics said its equipment has been on BETA aircraft that have flown more than 250,000 nautical miles.

Zipline starts Uber Eats drone delivery in the Houston area

Zipline said Uber Eats customers in Houston can select drone delivery, and The Verge reported on Nov. 17, 2026 that the first area is Spring, Texas, with Dallas next. Zipline says the aircraft fly up to 70 mph, carry up to 8 pounds for 10 miles, and lower orders from 330 feet. The company says it has completed 1 million drone deliveries in the Dallas metroplex and states a goal, with Uber, of 1 million deliveries a day by 2029.

Blue Origin will launch Google Suncatcher satellites

On Nov. 19, 2026 Blue Origin said it signed a launch services agreement to fly Google's Project Suncatcher satellites on New Glenn. The dedicated flight is targeted for the first quarter of 2029 and is meant to place solar-powered satellites, each carrying Google TPUs and about a data-center rack of computing capacity, into sun-synchronous orbit. Neither company stated a price or a satellite count. Google said the flight builds on work with other launch partners.

ChatGPT dictation shows text as you speak

On Nov. 18, 2026 OpenAI said Live Dictation is rolling out globally on every ChatGPT plan for iOS, Android, web, and the desktop app. Speech turns into text while the person is still talking, and the waveform can show or hide that text. OpenAI also said it removed the 10-minute recording limit. The note does not give a word-error rate for this release.

Anthropic shares open at $111.50

Anthropic Class A shares opened on Nasdaq on Nov. 19, 2026 at $111.50 under the ticker ANT, about 34 percent above the $83 price set on Nov. 18. CNBC, AP, and The Verge reported that opening print. AP still describes a $1.5 trillion valuation at the $83 price. Anthropic's Nov. 18 release expects closing on Nov. 20 and does not state an opening valuation or net proceeds.

OpenAI adds an Accounting plugin in ChatGPT

OpenAI's Nov. 18, 2026 release notes say an Accounting plugin can reconcile transactions, draft invoice entries, and prepare month-end reports from a user's records and templates. Members install it in the ChatGPT desktop app; Business workspaces do not need prior admin approval, while Enterprise workspaces do. Xero said on Nov. 19 that its month-end close review in this capability is available in the United States.

Nvidia says licensed Hopper shipments to China have stopped

On Nvidia's Nov. 17, 2026 earnings call, interim CEO Colette Kress said licensed Hopper data-center sales to China were $2.4 billion in the quarter ended Oct. 25, that the outlook includes none of those products, and that Nvidia is no longer shipping them to China. She did not give a reason or say whether the halt is permanent. The day's earnings release said the outlook assumes no Data Center compute revenue from China, without stating that shipments had stopped.

Google Cloud tools are built into Elementl nuclear planning

Elementl Power said on Nov. 17, 2026 that Google Cloud's Gemini for business and BigQuery will underpin Elementl Lattice, a planning suite whose first tool, Atlas, screens nuclear site characteristics. Google Cloud's Raiford Smith is quoted in the company release, which does not state a new power purchase or ownership stake. Data Center Dynamics reports that the companies already had a May agreement covering three US projects, early-stage development capital, and an offtake option, and that those project details were not disclosed.

Google opens Genie 4 to AI Ultra subscribers

Google said on Nov. 18 that eligible Google AI Ultra subscribers can use Genie 4 in Project Genie, including mid-session text edits and real-time native audio. The same-day model card describes 720p video at up to 24 frames per second and says visual details, object permanence, and audio can still drift. Google AI Pro members can join a waitlist. The card still says distribution through Project Genie is planned, and the post does not say a Gemini API path is open.

NASA and SpaceX move the CRS-35 return to Friday

NASA's media advisory, updated Nov. 18, now sets the CRS-35 Dragon undocking for no earlier than Friday, Nov. 20, at 12:05 p.m. EST, with splashdown off California no earlier than 5:04 p.m. EST on Nov. 21. SpaceX's mission page lists the same Friday undock time. Neither account describes a spacecraft fault. An earlier plan had pointed to Thursday.

Pony.ai reports Q3 revenue of $44.6 million

Pony.ai reported unaudited third-quarter revenue of $44.6 million on Nov. 18, up 75.2% from a year earlier. Robotaxi services were $17.8 million and robotruck services were $18.0 million. GAAP net loss widened to $136.3 million, which the company attributed primarily to a $77.0 million fair-value loss on trading securities.

Waymo pauses Bay Area rides during flooding

Waymo paused Bay Area robotaxi service on Nov. 18 after flooding. The app said rides would return when roads improved and the forecast was clear. A spokesperson told CBS the company had reduced the fleet before the pause, which was still in effect at 3 p.m. ABC7 also reported stalled Waymo vehicles in San Francisco. This is separate from the California freeway restriction, and neither report gave a restart time.

OpenAI agrees to surface Le Figaro in ChatGPT

OpenAI said on Nov. 18, 2026 that a partnership will make current and archive journalism from Le Figaro and associated publications discoverable in ChatGPT as summaries and excerpts with links to the original stories. OpenAI said Le Figaro keeps editorial control and plans to introduce ChatGPT Enterprise for its teams. The announcement states no price and no start date.

Musk says Anthropic may want 10 more gigawatts from SpaceX

At the New York Times DealBook Summit on Nov. 18, 2026, Elon Musk said Anthropic told him its highest-probability outcome is to keep renting SpaceX compute and to seek an incremental 10 gigawatts, and that Anthropic can end the contracts on 3 to 6 months' notice. He said a SpaceX-Tesla combination would need shareholder votes that may or may not succeed, and that he should not receive Tesla performance awards merely for completing a merger. Anthropic has not confirmed the 10-gigawatt request.

Microsoft makes Decision 1.1 generally available

Microsoft said on Nov. 17, 2026 that Microsoft-Decision-1.1 is generally available in Foundry and through OpenRouter and Vercel. The company said it is a Gemma 4 26B mixture-of-experts model that accepts images and is more than twice as fast as Microsoft-Decision-1, with Foundry input at $0.042 per million tokens on Global Standard, $0.046 in the US data zone, and $0.05 in the EU and APAC data zones, and with output tokens free. Microsoft said more than 4,000 customers have used it since a public preview, and its Nov. 18 Foundry post reports 78.8 percent average accuracy on 102 company benchmarks.

Google opens a waitlist for Stitch Loop

On Nov. 18, 2026, Google Labs introduced Stitch Loop, an experiment that connects to a GitHub codebase and live product data, writes code in an isolated cloud sandbox, and opens a pull request a person must approve before anything merges. Google said the default loops cover cybersecurity, latency, and accessibility, that the security loop uses Gemini 4 Carbon with DeepMind's CodeMender, and that teams can connect more than 180 integrations. The waitlist is open globally, with priority for eligible Google AI Ultra users, and Google stated no price.

Microsoft says it will use Intel EMIB for Azure silicon

Microsoft and Intel said they expanded a long-term agreement under which Microsoft is the first company to announce use of Intel Foundry EMIB packaging for Azure first-party silicon, with multi-year Xeon supply that includes Diamond Rapids CPUs. The releases do not name the Azure chip, a process node, a volume, a price, or a delivery date. Microsoft's page is dated Nov. 18, 2026 and its dateline reads Nov. 17; Intel's copy is dated Nov. 18. Performance multiples in the release compare differently configured systems and are not repeated here.

Google introduces Gemini 4 Neon

On Nov. 18, 2026 Google said Gemini 4 Neon is rolling out in the paid Gemini API, Gemini for business, Google Antigravity for Pro and Ultra subscribers, and the Gemini app for Pro and Ultra subscribers, at $3 per million input tokens and $15 per million output tokens with a 1 million-token context window. DeepMind's model card says Neon is adapted from Gemini 4 Carbon and puts the knowledge cutoff at March 2026. Google said Fairwind cyber access starts over the following days and that the full safety report is not yet published.

Anthropic prices its IPO at $83 a share

On Nov. 18, 2026 Anthropic said it priced 980,000,000 Class A shares at $83.00, with 956,351,354 sold by the company and 23,648,646 by selling stockholders. Anthropic said it receives no proceeds from the stockholder shares, that the SEC declared the registration effective the same day, that Nasdaq trading under ANT is expected on Nov. 19, and that closing is expected on Nov. 20, with a 30-day option for up to 147,000,000 more shares. AP reported a $1.5 trillion valuation; the company release does not state a valuation, net proceeds, or whether the private placement closed.

A Long March 2D orbits Shijian-24

Xinhua said a Long March-2D lifted off from Taiyuan at 6:51 a.m. Beijing time on Nov. 19, 2026 and placed Shijian-24 into its planned orbit. Xinhua called the launch a success and the 678th Long March flight. CGTN reported the same time, vehicle, site, and flight number. Neither report states the satellite's purpose or its orbit, and CGTN's 1.3-tonne, 700-kilometer line describes the rocket, not this mission.

Nvidia states up to $118 billion of guarantee exposure

Nvidia's Q3 fiscal 2027 CFO commentary says maximum gross guarantee exposure was $118.0 billion as of Oct. 25, 2026: $105.0 billion for SB Energy, $9.5 billion for power and systems, and $3.5 billion for neocloud land, power, and shell. The SB Energy guarantees cover about 4.25 GW across nine PORTS-Pike buildings in Ohio that Nvidia says will host its infrastructure on 20-year leases to OpenAI, and each guarantee starts only when that building is ready, with the first expected in fiscal 2029. Nvidia said it had made no guarantee payments and holds an option to support about 3.8 GW more.

/ Wednesday, November 18, 2026

Meta connects Loblaw checkout to Muse in Canada

On Nov. 18, 2026 Meta said Loblaw is the first Canadian retailer connected to Muse, so shoppers can plan meals, build a basket and check out with participating banners inside Muse. The same post says Muse voice is rolling out in Canada on mobile, web and Mac, and that Muse asks before a purchase while keeping passwords and payment methods out of the agent's view. Meta stated no price and no store count.

OpenAI says its models changed an AMD GPU sent to manufacturing

On Nov. 18, 2026 OpenAI said its models found up to a 14 percent power, performance and area gain versus heavily optimized baselines on AMD's flagship GPU design by rewriting circuit code, and that AMD included some of those changes in a design sent for manufacturing. OpenAI said AMD verified single-digit percentage gains on key blocks of an upcoming flagship CPU, and that model-proposed changes cut standard-cell area by about 13 percent in Jalapeno's vector unit and 10 percent in its matrix unit during silicon bring-up. Those figures are OpenAI's account, and the same post says OpenAI is training a GPT-Synopsys model with Synopsys.

AWS runs GPU instances inside Lambda

On Nov. 18, 2026 AWS said Lambda Event Functions on Lambda Managed Instances can run on EC2 G-series GPUs, with base images that include vLLM, SGLang and PyTorch. The AWS news blog lists G4dn, G5, G6, G6f, Gr6, Gr6f and G6e at launch, 3 to 48 GB of accelerator memory, and a timeout of up to 90 minutes for asynchronous and event source mapping invocations. Customers pay the EC2 on-demand price, plus a 9.75 percent management fee and $0.20 per million requests, in commercial regions where Lambda Managed Instances and those instances are available.

Nvidia puts RTX Spark compact desktops on sale

On Nov. 18, 2026 Nvidia said compact RTX Spark desktops from ASUS, Dell and MSI are available, with a 20-core Grace CPU, a Blackwell RTX GPU with 6,144 CUDA cores, up to 128GB of unified memory and up to 1 petaflop of AI performance. Nvidia says the systems can run 120-billion-parameter models locally. The Verge reported that the ASUS ProArt GR1X starts at $4,499 with 64GB of memory and 1TB of storage, while Acer, GIGABYTE, HP and Microsoft's Surface RTX Spark Dev Box were still described as coming.

Unauthorized Microsoft agent-framework releases are pulled from PyPI

PyPI's advisory says unauthorized copies of Microsoft's agent-framework package, versions 1.26.1 and 1.26.2, were uploaded on Nov. 17, 2026 with credential-stealing code and have been removed. Version 1.26.0 is not affected, and the advisory says PyPI Security worked with Microsoft on the remediation. Wiz, which reported the uploads, said the two versions appeared at 20:35 UTC and about 21:08 UTC and were unavailable by about 21:30 UTC. Wiz's Nov. 18 review of the captured requests counted three unique cases in which victim data left the host.

NuScale expands Everstar's Gordian across reactor-deployment work

On Nov. 18, 2026 NuScale said it is deploying Everstar's Gordian software across licensing, plant services, project management, supply chain, quality assurance, and product development for its NRC-certified small modular reactor. The company said a pilot produced efficiency gains and that the tool applies nuclear regulatory frameworks across connected documents. The release does not give a measured time or cost saving, a customer project, or a contract value.

Apple says an internal update caused Tuesday's service outage

Apple told NBC News and Reuters on Nov. 17, 2026 that an internal update caused outages and that the incident was not a cyberattack. NBC reported that Apple's status page had listed the App Store, Apple Music, AppleCare, iCloud Mail, and related business tools, and that the page showed no outages as of 6 p.m. ET. Reuters reported that iCloud and Apple Pay were among the affected customer services.

ULA says vehicle fuel supply, not the BE-4s, cut Vulcan thrust

On Nov. 18, 2026 United Launch Alliance said an initial review of Vulcan Leo 1 data indicates that low booster thrust came from inadequate fuel supply from the vehicle to the BE-4 engines, not from the performance of the engines. ULA said the cause is not yet determined. It repeated that the solid rockets, including the nozzle changes, performed as planned, and that the Centaur upper stage and its RL10 engines started normally but could not make up the lost performance. No return-to-flight date was given.

NASA and SpaceX delay the CRS-35 Dragon return to Thursday

NASA and SpaceX stood down from the Wednesday, Nov. 18, 2026 undocking of the CRS-35 cargo Dragon because of weather at the California splashdown sites, and both NASA's station blog and SpaceX's mission page list the next opportunity as Thursday, Nov. 19, at 12:05 p.m. ET. SpaceX says this Dragon, on its fourth flight, has been docked for 35 days. Neither account describes a spacecraft fault.

Boston Dynamics plans Atlas tests at GlobalFoundries

Boston Dynamics said on Nov. 17 that GlobalFoundries will expand Spot robots across fabs in Malta, New York; Burlington, Vermont; Dresden; and Singapore, and will be the first semiconductor manufacturer in its Atlas early-adopter program. Onsite Atlas testing is planned for early 2027, with no robot count, price, or purchase order disclosed. The release's Singapore figures, nearly 900 kilometers and more than 90,000 inspections, describe Spot robots already working there.

Google makes Gemini Tasks free in six countries

Google said on Nov. 17 that Tasks in the Gemini app is rolling out free to hundreds of millions of people in the United States, Japan, Korea, Canada, Australia, and India, using Gemini 4 Hydrogen. In the United States, Tasks can call businesses, take requests by text or by email to gemini@google.com, and check out with Google Pay after the user confirms. Calling is not available in Indiana, Louisiana, Minnesota, Montana, or Nebraska, and Google said American Express support is coming soon.

ICEYE offers EUR 41 a share for all of Avio

ICEYE said on Nov. 17 that its Italian subsidiary Iris 2 will offer EUR 41 in cash for each Avio ordinary share, a 21.3 percent premium to the Nov. 16 close. The Italian Article 102 notice sets a maximum outlay of EUR 2,052,657,251 for up to 50,064,811 existing shares, excluding later incentive shares, and requires a stake of at least 66.67 percent. Avio's board said it has started assessing the offer, and ICEYE's additional EUR 1 billion plan for Italy is described as following completion and subject to the offer conditions.

Google will keep some Gemini chats for about two months

Google said on Nov. 16 that Gemini users will be able to keep chat history while opting out of generative AI training, and that future chats made with the new History setting off will be saved for about two months. The change is set to roll out globally in the Gemini app over the coming weeks. TechCrunch, citing a privacy-hub archive from Nov. 12, said the previous retention for those chats was 72 hours, and said the updated hub states that temporary chats are not used to train models.

Microsoft opens Salesforce in Copilot in public preview

Microsoft and Salesforce said on Nov. 17 that Salesforce in Copilot is in public preview with more than 38 prebuilt sales skills, and that querying Fabric OneLake data from Salesforce Data 360 is generally available. Sharing Data 360 data into OneLake is in public preview, and Sales Cloud in Teams is in a customer pilot. Accenture is the design partner and a pilot customer, and the release states no price.

Firefly signs QASCOM to fly COLLINS on a lunar Elytra

Firefly said on Nov. 17 that it signed a commercial agreement to host QASCOM's COLLINS navigation payload on Elytra in lunar orbit on the Blue Ghost mission to the Gruithuisen Domes, targeted for no earlier than 2028. The mission now lists eight payloads, six of them NASA CLPS payloads. Elytra is to relay for the lander, run COLLINS for six months with an option to extend, and then remain in orbit for five years for Ocula imaging, and Firefly stated no price.

POSCO agrees to supply CATL with lithium through 2028

POSCO said on Nov. 18 that it signed an agreement to supply CATL with lithium for electric-vehicle batteries from 2026 through 2028, and that deliveries have already started. POSCO did not disclose the volume or the contract value. The Korea Herald reported that the lithium's source and destination market were also withheld, and no matching CATL announcement was found.

Tesla unveils the 7.4 MW Powercore power unit

Tesla introduced Powercore on Nov. 17, a single enclosure that combines a solid-state medium-voltage transformer, an inverter, and a breaker, rated at 7.4 MW (9 MVA) continuously at 34.5 kV AC. The company says it will be built in the United States and is planned to operate in 2027, for solar plants, battery storage, and data centers that take 800 V direct current. Tesla told pv magazine it is working with Nvidia to certify Powercore for the DSX Ready program, and it named no price, customer, or factory.

ULA says low booster performance kept Vulcan Leo 1 from orbit

ULA said on Nov. 17 that early data from Vulcan Leo 1 show low booster performance just after the Nov. 15 liftoff, so the stages separated too low and too slow to reach orbit. Solid-rocket performance, including the GEM 63XL nozzle changes, was nominal, while Centaur V and its RL10 engines started as planned and could not recover the mission. CEO Mark Peller told SpaceNews the probe is focused on the liquid-fueled core stage; the FAA is overseeing it, ULA named no failed part, and it set no return-to-flight date.

Nvidia expands substrate and packaging supply agreements

Nvidia said on Nov. 17, 2026 that Kinsus plans to expand advanced ABF substrate capacity dedicated to Nvidia by more than 3.5 times at its Youth campus in Taoyuan, with production expected to begin in the second half of 2027 and additional capacity targeted for the third quarter of 2028. Samsung Electro-Mechanics amended an existing substrate agreement covering factories in Sejong and Busan so those plants can support higher production. A multiyear SPIL agreement, backed by multibillion-dollar funding from Nvidia, is meant to increase contracted advanced packaging capacity dedicated to Nvidia by mid-2029.

Microsoft opens Copilot Workspaces and starts Autopilot rollout

Microsoft said on Nov. 17, 2026 that Copilot Home, where Chat and Cowork come together, is generally available, and that Copilot Code is due in the coming weeks. Copilot Workspaces, a shared space for people and agents, is available in Frontier, and Autopilot, an agent with its own Entra identity and a named manager, has started rolling out to Frontier customers. Apps built with Copilot Code are hosted in the Copilot Managed Runtime, which is generally available.

Anthropic filing sets post-IPO stakes and a $26 billion placement

Anthropic's Nov. 17, 2026 amendment to its Form S-1 still estimates an IPO price of $70 to $78. After the offering and a concurrent private placement, and using Sept. 30, 2026 share counts after the conversions named in the prospectus, Amazon would hold 14.7 percent of outstanding capital stock and Google LLC 13.7 percent, both in Class N shares that have no vote except as required by law. Seven companies agreed to buy $26 billion of Class N stock at the IPO price; at an assumed $74, that is about $10 billion each for Google and Nvidia, $2.5 billion for AMD, $1 billion each for Broadcom, Micron, and Microsoft, and $500 million for TSMC Partners, and the sale closes only if the IPO is completed.

Nvidia reports $115.8 billion in fiscal Q3 revenue

Nvidia reported unaudited revenue of $115.844 billion for the quarter ended Oct. 25, 2026, up 20 percent from the prior quarter and 103 percent from a year earlier. GAAP net income was $65.287 billion, or $2.70 a diluted share, and the release puts Data Center revenue at $107.6 billion. It guided the next quarter to $130.0 billion, plus or minus 3 percent, with gross margin of 71 to 72 percent and no assumed Data Center compute revenue from China.

OpenAI releases GPT-6.2 Astra and Sol

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Nov. 17, 2026 in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, and in the API. Astra is $8 per million input tokens, $0.40 cached, and $40 output; Sol is $2, $0.10, and $10. OpenAI says both meet its Critical cybersecurity threshold and its High biological and chemical threshold, and that Astra is its first external model rated High for AI self-improvement.

Microsoft releases MAI Code 2 and Code 2 Flash

Microsoft released MAI Code 2 and MAI Code 2 Flash on Nov. 17, 2026. GitHub says Code 2 is in Copilot for Pro, Pro+, Max, Business, and Enterprise annual subscribers, and Code 2 Flash is also on Free and Student annual plans, with usage-based billing and no per-token price in the posts. Microsoft says the models target everyday enterprise coding and are not yet tuned for fundamental science research or very long-horizon agentic work.

Nvidia's board says it does not intend to reinstate Huang

On Nvidia's Nov. 17 earnings call, board chair Harvey C. Jones said the board unanimously accepted Jensen Huang's resignation and does not intend to reinstate him as CEO. Jones said the board has retained Heidrick & Struggles and will consider internal and external candidates without installing a permanent CEO before it finds the right person. Colette Kress remains interim CEO and said that, from a business perspective, it is business as usual.

/ Tuesday, November 17, 2026

Sources say Anthropic may price its IPO above $70 to $78

CNBC reported on Nov. 17 that sources said Anthropic is likely to price its IPO above the filed $70 to $78 range and plans to close the order books at 4 p.m. ET that day. Bloomberg, as reported by Quartz, said guidance is to the top of the range or above, that orders are almost seven times the shares on offer, and that pricing is planned for Wednesday with trading the next day. Anthropic is not quoted, and both accounts say the terms can still change.

AMD says the Taalas acquisition is complete

AMD said on Nov. 17 that it completed the acquisition of Taalas, a specialist in AI inference silicon. The newsroom says AMD will put the technology into its accelerator roadmap and into systems alongside Instinct GPUs, with EPYC, ROCm, and Helios, and that the deal expands engineering in Canada. The release does not state a price, a share count, or a product date.

House sets Dec. 1 and Dec. 2 interviews with Huang and Nvidia's counsel

The chairs of the House Select Committee on China, Foreign Affairs, and Intelligence said on Nov. 17 that Jensen Huang agreed to a transcribed interview on Dec. 1 and that Nvidia general counsel Tim Teter agreed to one on Dec. 2. They said both appearances are voluntary. An Nvidia spokesperson told NBC the company looks forward to meeting the chairs on Dec. 2 and did not mention Huang's date.

California DMV bars Waymo from driverless freeway driving

California's DMV said a restriction that took effect Nov. 11 bars Waymo from driverless freeway driving and testing in the state until the company shows it can handle construction zones and emergency vehicles. The department cited three vehicles that entered a U.S. 101 construction closure near Burlingame and a separate Highway 280 encounter with the California Highway Patrol. Waymo said those 101 encounters caused no injuries or collisions, and that it has also paused freeway rides in its other markets while surface-street service continues.

Huang says he would return as Nvidia CEO if the board asked

Jensen Huang said on Nov. 17 that he would be honored to return as Nvidia CEO if the board decided a return served shareholders, employees, the company, and America. In the written statement, obtained by CNBC and reported by NBC, he said he had made serious personal mistakes and would cooperate with the board's independent investigation. CNBC reported Monday that sources said the board was not considering his return. Colette Kress remains interim CEO.

Tesla plans a €3 billion Berlin hall for Cybercab output in 2029

Tesla said on Nov. 16 that it will spend about €3 billion on a new manufacturing building at Giga Berlin in Gruenheide for Cybercab production, with capacity of up to 1 million vehicles a year. Plant director Andre Thierig said construction is expected to start in the first half of 2027 and production in 2029. Electrek, citing Tesla, said the sum also covers ancillary buildings and a second wastewater plant, and that Tesla plans to file the building application before the end of 2026.

Qualcomm opens Dragonfly AI200 to select Cirrascale evaluations

Qualcomm said on Nov. 17 that Dragonfly AI200, its rack-scale inference system, is available for evaluation by select customers on Cirrascale Cloud Services. The company says each card has 768 GB of LPDDR memory and that the rack uses direct liquid cooling, PCIe scale-up and Ethernet scale-out. A demonstration is scheduled at SC26 in Chicago on Nov. 17-18. Qualcomm describes this as evaluation access, and the release does not state a general production shipment.

OpenAI signs a Chile AI memorandum with no purchase commitment

OpenAI said on Nov. 17 that it signed a memorandum with Chile's Ministry of Science, Technology, Knowledge and Innovation to explore AI use in public administration, civil-service training and startup support. Minister Ximena Lincolao signed for Chile at Semafor's Silicon Valley & The World conference in California. OpenAI says the memorandum sets no investment or purchasing commitment, and that specific activities will be agreed separately.

Space Force modifies NSSL Lane 2 contracts, ceiling $985 million

The Nov. 16 Department of War contract list awards SpaceX, United Launch Services and Blue Origin modifications to their National Security Space Launch Phase 3 Lane 2 contracts, with a combined maximum ceiling of $985 million. The notice obligates no funds at award. Work is listed at the contractors' sites, Cape Canaveral, Kennedy Space Center and Vandenberg, and is expected to be completed by April 2033. The notice does not split the ceiling by company or state how many missions it adds.

Huawei Cloud announces the CloudRobo-WM1.2 world model

Huawei Cloud held an online event on Nov. 16 whose agenda listed the official release and open-sourcing of CloudRobo-WM1.2, which it calls its first Ascend-native embodied world model. Cailian reported that product chief Wu Xiaodong said the company will open-source the weights, inference and code, that Huawei said the model ranked first on RoboTwin 2.0 and LiberoPro 2.0, and that the CloudRobo platform has more than 16,500 developers. The pages checked did not show a public download or license.

Nvidia's board adds an export-control oversight committee

On Nov. 17, 2026, Nvidia said its board created a standing National Security and Export Control Oversight Committee. Director Suzanne Nora Johnson will chair it, with Stephen C. Neal and John O. Dabiri. The committee will meet at least quarterly. Nvidia said its national-security and global-trade teams will report to the committee and to interim CEO Colette Kress.

US agencies detail a 2024 Megaspeed chip investigation

Commerce and Justice said Commerce opened an export-control investigation of Singapore-based Megaspeed in 2024. They say Commerce asked Nvidia in May 2025 for records of direct or indirect sales, and that Nvidia placed Megaspeed on a no-sell list in June 2025. That month, Bureau of Industry and Security agents and Singapore police did an on-site check and interviewed then-CEO Alice Huang. Justice says it then took the lead in a criminal investigation. Nvidia is not quoted in the statement.

Isomorphic Labs raises a $5 billion Series C

Isomorphic Labs said on Nov. 17, 2026 that it raised $5 billion in a Series C co-led by Greenoaks and Sequoia Capital, with Andreessen Horowitz and other new investors, plus continued significant investment from Alphabet. Thrive Capital, MGX, GV, CapitalG, Temasek, and the UK Sovereign AI Fund also participated. The company said the round will scale its IsoDDE drug-design engine and candidate pipeline. The announcement does not state a valuation or Alphabet's check size.

Google agrees to buy 50 MW of Oregon geothermal power

Quaise Energy said on Nov. 17, 2026 that Google signed a power-purchase agreement for 50 MW from Project Obsidian in central Oregon, with purchases beginning as early as 2030 for data centers in The Dalles. The first phase uses conventional drilling, and a later phase is planned to use millimeter-wave drilling. OPB, after interviews with both companies, says this is Quaise's first commercial geothermal agreement and that the price was not disclosed.

OpenAI says NATO uses up to 300 billion tokens a month

OpenAI said on Nov. 17, 2026 that it works with NATO's Communications and Information Agency to give NATO's core institutions access to cybersecurity models through Daybreak. The company says NATO has used up to 300 billion raw input and output tokens per month. It also said participants at the 2026 Crossed Swords exercise built and evaluated Daybreak agents with human oversight. No start date or price is stated.

First permanently based F-15EXs land at Kadena

Two Boeing F-15EX Eagle IIs assigned to the 67th Fighter Squadron landed at Kadena Air Base in Okinawa on 15 Nov 2026, the first permanently assigned F-15EXs at a U.S. Air Force base outside the United States. The 18th Wing said Kadena is the second site to field the jet, after the Oregon Air National Guard, and that the 44th and 67th Fighter Squadrons will operate it while the base replaces its F-15C/D fleet. The Air Force plans to station 36 F-15EXs at Kadena, and a wing spokeswoman said further arrivals depend on production and have no public schedule.

Li Auto says it will sell Mach M100 chips outside the company

Li Auto CEO Li Xiang wrote on Weibo on 17 Nov 2026 that several robotics companies had asked to buy the Mach M100 and that Li Auto will gradually open external supply of the chip. He named no customers, announced no purchase agreements, and gave no start date. He also said the 5-nanometer chip, rated at 1,280 TOPS, is already standard on several new Li Auto models.

UK will end Palantir's NHS contract in 2029

Health minister James Frith told MPs on 16 Nov 2026 that the NHS England federated data platform contract with Palantir will stay in place for two more years and end in 2029, rather than using a break clause that could have ended it in spring 2027. He said a replacement procurement will start in the second half of 2027, take about 18 months, and require interoperability and common standards, and he did not say whether Palantir can bid. Palantir said the decision was good news for patients and that it would deepen its UK investment.

California sues OpenAI over model-testing safeguards

On Nov. 16, 2026, California Attorney General Rob Bonta sued OpenAI Foundation and OpenAI Group PBC in San Francisco Superior Court, alleging a public nuisance and an Unfair Competition Law violation over testing of advanced models, and seeking an injunction, restitution, civil penalties of $2,500 per violation, disgorgement, and abatement. The Attorney General says the July Hugging Face attack lasted more than two days and was one of at least 80 incidents; the posted complaint shows no case number. OpenAI told the BBC the suit singles out a challenge identified by every major lab, and that it will keep engaging with the office.

Nvidia opens Nemotron chemistry data and science recipes

On Nov. 16, 2026, Nvidia introduced Nemotron for Science, packaging the already released Nemotron 3.5 Super VL model with post-training recipes and LitmusBench v0.2, an open chemistry dataset of nearly 50,000 questions. Nvidia said both are available on Hugging Face and that the model was optimized for a 128GB DGX Spark. Company-reported results include about 76 percent of annotated reactions recovered on RxnScribe, 61 percent accuracy on a quantum-dot protocol versus 54 percent for GPT-6 Astra, and an 81 percent repair rate on a GROMACS test.

Instagram tests Direct, a standalone messaging app

On Nov. 16, 2026, Meta said Instagram is testing Direct, a standalone messaging app, in Turkiye and Thailand on Android and iOS. People can sign up with an email or phone number, or use an Instagram login so existing chats sync; without an Instagram account, only contacts can start a chat. The newsroom post also prints Nov. 13, 2026. Users must be at least 13, and Meta has not given a user count or a date for other countries.

OpenAI offers a free classifier for child-abuse images

On Nov. 16, 2026, OpenAI said it released an image classifier that scores possible child sexual abuse material, including a signal for AI-generated images, and is offering it at no cost through the API to vetted organizations. OpenAI said NCMEC assisted with training and evaluation, and that in its own comparison the classifier would flag about 48 percent fewer user-uploaded images while finding roughly the same amount of abusive material, with flagged images about twice as likely to contain abuse. OpenAI also said ChatGPT warnings will remove the flagged conversation and pause access for 24 hours.

Nvidia announces Rubin HPC, expected in 2028

At SC26 on 16 Nov 2026, Nvidia announced Rubin HPC, a Vera Rubin-platform GPU for simulation and data processing in science, engineering, and analytics. Nvidia says the chip has more than 50,000 CUDA cores, the most on any of its GPUs, plus sparse-data acceleration and seventh-generation NVLink, and that it is expected in 2028. Nvidia did not disclose memory capacity or FP64 throughput.

MLCommons releases Training v6.1 with first Rubin results

MLCommons released MLPerf Training v6.1 on 16 Nov 2026, with 153 systems and 250 results, and said the round included Nvidia Rubin-generation GPUs and Google's TPUv7, called Ironwood, for the first time. The results explorer lists Nvidia's Hecate Rubin system at 3.947 minutes on Llama 3.1 8B with 512 GPUs and a CoreWeave Rubin system at 7.162 minutes with 72 GPUs. Nvidia says Vera Rubin NVL72 delivered up to twice the performance per GPU of its GB300 NVL72.

XPeng reports Q3 revenue of RMB 24.28 billion

XPeng said unaudited third-quarter 2026 revenue was RMB 24.28 billion (US$3.62 billion), up 19.1 percent from a year earlier and 23.0 percent from the prior quarter, with 118,390 vehicle deliveries. Net loss was RMB 0.87 billion, wider than RMB 0.38 billion a year earlier, and cash was RMB 45.41 billion as of 30 Sep 2026. For the fourth quarter the company guided to 169,000-174,000 deliveries and revenue of RMB 30.40-31.80 billion.

OpenAI expands free ChatGPT access for researchers

OpenAI said on 16 Nov 2026 that it is expanding ChatGPT for Academic Researchers toward 200,000 researchers, double the goal it set when the July pilot launched. The next round opens 23 Nov and will award 50,000 grants on a first-come basis, each covering the recipient and up to two collaborators on ChatGPT Plus instead of the pilot's lottery and ChatGPT Pro seats. OpenAI said 20,000 researchers in 33 countries have joined since July, and that program data is not used for training by default.

LVMH media group signs an OpenAI agreement

On 16 Nov 2026, LVMH Group's Media Division and OpenAI said they signed an agreement to deploy OpenAI tools and ChatGPT Enterprise across the division and to show attributed summaries and excerpts of its journalism in ChatGPT, with links to the original articles. The division, Groupe Les Echos-Le Parisien, lists 17 media brands and says it has 1,800 employees, including nearly 700 journalists. Neither announcement states a price, a duration, or whether the articles may be used for model training.

/ Monday, November 16, 2026

Anthropic will cut high-severity bug disclosure to 30 days

Anthropic said on Nov. 16 that reports sent on or after Jan. 1, 2027 will have a 30-day public-disclosure window for high- and critical-severity vulnerabilities and 45 days for other issues, down from 90 days. It will aim to escalate after 15 days if a maintainer does not respond, and wait 30 days after a patch before publishing full technical details. The seven-day window for flaws under active exploitation stays in place, in-progress reports keep their deadlines, and findings with no human review, including OSS Scanner, are excluded. Maintainers of already overdue high- and critical-severity reports get another 30 days before publication starting Jan. 1, 2027.

Pony.ai starts fully driverless bZ4X robotaxi service in Guangzhou

Pony.ai said on Nov. 16 that its Gen-7 Toyota bZ4X robotaxi joined the fully driverless commercial fleet in Guangzhou and can be hailed 24 hours a day in the PonyPilot app and on WeChat and Alipay. The company said the first mass-produced vehicle came off the line in February and that it is working toward about 1,000 bZ4X robotaxis in major Chinese cities. CnEVPost, citing company statements, says the initial area is Nansha and that the companies gave no vehicle count or date for the 1,000-vehicle plan.

AMD closes the World Labs deal and names Fei-Fei Li chief scientist

AMD said on Nov. 16 that it completed the acquisition of World Labs and that Fei-Fei Li joined as executive vice president and chief scientist, reporting to Lisa Su. A Form 8-K/A filed the same day says the all-stock acquisition closed on Nov. 13 and that AMD issued 14,978,322 shares. The filing restates the September agreement price of about $8.2 billion, subject to adjustment, and does not state a final dollar value.

A source says Nvidia's board is not considering Huang's return

CNBC reported on Nov. 16 that a person familiar with the board said Nvidia is not considering bringing Jensen Huang back as CEO and is focused on the investigation. The Washington Post reported the same point from a person familiar with the board's thinking. Nvidia is not quoted confirming the account, and the two reports may be one source. Colette Kress remains interim CEO. CNBC also reported that Brad Gerstner said Huang had not discussed a comeback with him, and that President Trump said he is open to Huang returning.

OpenAI signs four firms for embedded model assessments

OpenAI said on Nov. 16 that it signed agreements with Apollo Research, METR, Transluce, and Goodfire for a first phase of embedded assessments of alignment and safeguards during training and internal use. Assessors can receive checkpoints, code, training data, and monitoring access, with on-site work for sensitive material, and OpenAI can redact confidential information before publication. Apollo and Goodfire are paid on terms that do not depend on the findings. METR and Transluce are not paid and may receive compute. A further METR monitor assessment is still pending agreement on scope, and a broader Transluce assessment of about eight weeks is described as a shared intent for early 2027.

Commerce releases a remote-access rule for advanced AI chips

On Nov. 16, 2026 the Commerce Department's Bureau of Industry and Security released interim final rule RIN 0694-AK50. The copy is on display and still has a blank Federal Register date. It requires a license for advanced computing items destined for data centers that would give companies headquartered in Country Group D:5 or Macau access to more than 12 million aggregate total processing performance, including across linked data centers, and it adds licenses for exports to Indonesia, Malaysia, Singapore, Thailand, and Brazil unless the recipient is U.S.-headquartered and the exporter files a new report. Commercial recommender systems are excluded, existing covered services have 90 days after the effective date to get a license, and the text formally rescinds the January 2025 AI Diffusion Rule.

OpenAI opens computer-use agents for healthcare workflows

OpenAI said on Nov. 16, 2026 that computer-use agents are available for healthcare workflows in ChatGPT for Healthcare, ChatGPT for Clinicians, and the API, and it named Abridge, Ambience, Cigna Healthcare, and CVS Health as partners. On OpenAI's review of GPT-6 Astra inbox tasks, 99 percent of 325 tasks were rated safe and physicians called 93 percent of 646 assessments very or extremely helpful. On Stanford's HealthAdminBench, OpenAI said Astra scored 94.3 percent of rubric points and Luna 91.8 percent at 8.5 cents a task, both at maximum reasoning effort.

TensorWave says it will supply OpenAI with 457 MW of AMD capacity

TensorWave said on Nov. 16, 2026 that it signed a six-year agreement to deliver 457 megawatts of dedicated, single-tenant AI infrastructure for OpenAI at multiple U.S. sites, using AMD Instinct MI455X GPUs in Helios racks linked by Ethernet. It expects first capacity in the first half of 2027 and the full deployment by early 2028, and it called itself the first publicly named neocloud under the existing OpenAI-AMD partnership. The post quotes AMD executive Vamsi Boppana, states no contract value, and does not quote OpenAI.

Google says attackers automated exploit writing without a person in the loop

Google's Threat Intelligence Group said on Nov. 16, 2026 that, in the past quarter, it identified threat actors using language-model frameworks to fully automate vulnerability discovery and exploitation without a human in the loop for the first time. GTIG described an exposed staging server running a framework it calls exploit_agent, which compares code before and after patches, drafts exploits, and runs them automatically. Telemetry on that server pointed to French media, entertainment, and education sites, and GTIG said it has also seen the open-source tool CAIRN used against U.S. aerospace targets.

Microsoft traces the 365 outage to a July gateway defect

The Register says a preliminary Microsoft report published Nov. 15 traces the Nov. 12 Microsoft 365 outage to a July defect in network gateway servers: lock contention exhausted CPU and memory, and the problem is worse on machines with more cores. A Teams failover onto the gateway that also serves Exchange Online, combined with almost 14 percent of gateway capacity being out, cascaded until more than 80 percent of servers were unresponsive. Expert Insights dates that preliminary report to Nov. 16, the admin center copy was not opened, and The Register says the final report is due Nov. 20.

Zoox opens Las Vegas robotaxi matches on the Uber app

Zoox said on Nov. 16 that Uber riders in Las Vegas can be matched with a Zoox robotaxi on eligible trips, and that rides remain available in the Zoox app. TechCrunch, citing the companies, says those trips are UberX, Uber Comfort, Uber Electric, Uber Comfort Electric, and Wait & Save, and that riders cannot select a Zoox directly at first. Zoox said the Uber fleet is starting small, that it and Uber are the official robotaxi and rideshare partners of MGM Resorts properties in Las Vegas, and that an Uber launch in Los Angeles will come next year, after Zoox-app service opens.

Baidu says Apollo Go reached 30 cities and 390 million kilometers

Baidu said in its Nov. 16 third-quarter results that Apollo Go was in 30 cities as of October, drove about 34 million autonomous kilometers in the quarter, and had accumulated more than 390 million autonomous kilometers by Oct. 31. Fully driverless operational rides in Shenzhen rose about 170 percent from the previous quarter, and Baidu said it doubled the Dubai fleet while broadening that fully driverless area. London testing with Uber and Lyft and fully driverless testing on Hong Kong's Airport Island continued, and Baidu says the operating figures are from unaudited internal accounts.

WSJ says the US plans to name AI national champions

The Wall Street Journal reported on Nov. 15, as relayed by Reuters, that the administration plans to designate more than two dozen AI national champions, including Nvidia, Amazon, Microsoft, OpenAI, and Anthropic, to help them expand overseas. Moneycontrol's account of that report says an announcement is expected within the next week, that the timing could change, and that Commerce Secretary Howard Lutnick is expected to name the companies. Selected firms could receive support for overseas projects, federal financing, and faster export-license reviews. Reuters said it could not immediately verify the report. No designation has been announced.

WSJ says Huang told cloud buyers to bring customer prepayments

The Wall Street Journal reported on Nov. 16 that at a Berlin conference last month Jensen Huang told emerging cloud companies they would need upfront customer payments to receive GPU allocations. The Journal says startups that relied on Nvidia financing now expect a further pullback under interim CEO Colette Kress. UBS analyst Tim Arcuri's estimate that a pullback could remove $50 billion or more is the analyst's, not Nvidia's. Nvidia is not quoted confirming a new rule. The same article says people familiar described a closed $3 billion Perplexity investment, an unfinished Reflection AI offer, and a Thinking Machines investment still being finalized at a $40 billion valuation. The Journal page was not opened; this uses Hindustan Times' copy.

Alibaba opens Qwen4-Max on the API at $1.50 per million tokens

Alibaba's Nov. 16, 2026 post says Qwen4-Max is available on QwenCloud through the Token Plan and the API. The company says the model has 2.4 trillion parameters, activates 71 billion per token, accepts images and video, and has a 1 million-token context. QwenCloud lists input at $1.50 and output at $4.50 per million tokens. On Alibaba's Work360 test, the company says it uses 35 percent fewer tokens than Qwen3.8-Max at high effort and cuts average task cost by about 60 percent. Smaller Qwen4 models are planned as open weights. The Nov. 11 preview remains a separate listing scheduled to retire on Nov. 25.

Nvidia says its charged Taiwan manager did not know of a diversion

Bloomberg reported on Nov. 15 that Nvidia told Taiwanese political officials it disagreed with charges against a manager surnamed Chang and believed he should not be under investigation. Nvidia told Bloomberg it investigated and found no indication the employee knew of a diversion by employees at other companies. Keelung prosecutors, who indicted Chang in August, said they were unaware of those government contacts. TSMC declined to get involved and declined to comment. The 74-server and 56-server figures are the prosecutors' account in Bloomberg's report. The Bloomberg page itself was not opened; this uses NDTV Profit's unedited syndicated copy.

Microsoft will replace Phi Silica with Aion Instruct

On Nov. 13, 2026 Microsoft said Aion Instruct will replace Phi Silica as the on-device language model for Windows. The blog lists Aion Instruct at 2.3 billion parameters with a 32K-token input limit, against Phi Silica at 3.8 billion parameters and 3.5K tokens, and says it will also run on supported Nvidia RTX and AMD RDNA4 GPUs. A developer package is already available, Insider Preview rollout is planned for Nov. 20, and retail availability is planned for January 2027, when Microsoft Learn says Phi Silica will be removed.

Apple opens a US study of how people use health AI

On Nov. 15, 2026 Apple opened the Health AI Usage Study in the United States for up to 200,000 people age 21 and older who have an iPhone 15 Pro or later running iOS 26 or later. The study app, built with OpenEvidence, answers health questions with peer-reviewed citations and lets participants choose whether to share HealthKit data such as heart rate, activity, sleep, or lab results. Apple says the app does not diagnose or treat conditions and that study data will not be used to train AI models.

FT says Huang has discussed returning as Nvidia CEO

The Financial Times reported on Nov. 15, 2026 that Jensen Huang has discussed returning as Nvidia's chief executive with confidants, and that an ally said supporters were lobbying for his reinstatement, including through the White House, with his backing. CNBC's account of that report says Nvidia and Huang declined to comment to the FT. A spokesperson told CNBC that the board is selecting a search firm, has not formally engaged any candidate, and that talk of named candidates is premature.

Bessent says Nvidia should hire its next CEO from outside

On Nov. 15, 2026 Treasury Secretary Scott Bessent said on Fox News that Nvidia's board should hire its next CEO from outside the company and that he wants answers within 10 days from the board and outside counsel. CNBC and Fox Business both reported the interview. Colette Kress remains interim CEO, and Bessent's remarks are not a finding that Nvidia broke export rules.

Bloomberg says Apple still targets 2027 for first smart glasses

Bloomberg's Mark Gurman wrote on Nov. 15, 2026 that Apple still plans to show its first smart glasses at WWDC in June 2027 and ship them before the end of that year, while saying software delays could push both dates. MacRumors' account of the newsletter says the prototypes, code-named N50, have no display, rely on a paired iPhone, and use Siri for tasks in existing iPhone apps. Apple has not announced the product, and no price was reported.

SK hynix sets a May 2027 opening for its Yongin test line

SK hynix told reporters on a Nov. 10 tour of its Yongin site, in accounts published Nov. 16, that the Trinity Fab cleanroom will open in May 2027 and that supplier evaluations will start in the fourth quarter. The 874.6 billion won project includes 373.6 billion won from the central government and 57.7 billion won from Gyeonggi and Yongin, and Chosun Biz describes SK hynix's roughly 443 billion won share as cleanroom space and staff. The test line is planned around argon-fluoride immersion tools, not EUV, and an independent foundation will operate it.

/ Sunday, November 15, 2026

Vulcan fails to orbit on its first Amazon Leo flight

ULA's Vulcan lifted off from Cape Canaveral's SLC-41 at 12:40 p.m. EST on Nov. 15 with 40 Amazon Leo satellites and failed to reach orbit after an anomaly about 10 minutes into the flight. ULA said it did not yet have specific data on the cause and would investigate with its customer before returning to flight. Spaceflight Now called it ULA's first outright mission failure since 2006 and Vulcan's first flight in nine months.

A judge lets the DRAM supply case against Samsung and SK hynix proceed

On Nov. 14, Judge Noel Wise of the Northern District of California denied in full a motion to dismiss a proposed class action accusing Samsung, SK hynix, and Micron of conspiring to restrict conventional DRAM supply. The docket records the order in Garciaguirre v. Samsung Electronics, and LitNews reports the court found a Sherman Act claim plausible while treating the allegations as unproven. Answers are due Dec. 7, 2026, with a case conference set for Jan. 12, 2027.

Philippine Airlines says it signed for 15 Boeing 787-10s

Philippine Airlines said on Nov. 15 that it signed agreements with Boeing and GE Aerospace on Nov. 10 for 15 Boeing 787-10 jets, plus an opportunity to buy five more, with GEnx-1B engines and delivery in the next decade. Boeing sales chief Brad McMullen is quoted in the airline's release. Boeing's July 20 release had described the same package as a commitment once finalized, and no price was stated.

Axiom closes the design review for its NASA spacesuit

Axiom Space said on Nov. 13, 2026 that it closed the critical design review for the AxEMU spacesuit it is building for NASA, after reviewers examined component evidence, drawings, analyses and prototype tests, including whole-suit mass and pressure drop. The company said the suit now moves into production and qualification testing, with pressure-garment and portable life-support assembly targeted for early 2027 ahead of an International Space Station demonstration. A Sortie configuration for early Artemis lunar excursions builds on those AxEMU components.

New York qualifies Holtec, TerraPower and X-energy for nuclear bids

The New York Power Authority said on Nov. 12, 2026 that it qualified 10 developers for a planned advanced-nuclear project of at least 1 gigawatt in upstate New York. Holtec, TerraPower and X-energy advanced only on the small-modular-reactor path, while six other firms qualified for both large reactors and small modular reactors. NYPA said the qualified firms will be invited to later solicitations, and that a host community, technology and development partner are still to be chosen.

Carbon County approves a permit for Valar's Raceway campus

Carbon County's planning commission on Nov. 12, 2026 approved a conditional use permit for Valar Atomics' proposed Raceway Energy Campus on about 640 acres of Utah state trust land, with Commissioner Lisa Simmons voting no. The permit sets conditions for natural-gas power, data centers, fuel work and later reactors, including industrial water, emergency planning and federal licenses before nuclear construction or operation. The Salt Lake Tribune reported the company's figures of about 3 gigawatts and 40 to 120 reactors, and that a gas plant could still be pursued if nuclear licenses are not issued.

Sungrow launches the PowerStack 3.0 storage series

Sungrow said it launched the PowerStack 3.0 commercial and industrial storage series on Nov. 11, 2026 at its Global Energy Pioneer Summit in Hefei. The company said one cabinet dock accepts a power-conversion, solar, charging or microgrid module, and that the microgrid module can switch to off-grid power without a separate transfer switch. Sungrow stated an energy density of up to 256 kilowatt-hours per square meter, converter efficiency of 99.2 percent and round-trip efficiency above 92 percent with a 684 ampere-hour cell.

Musk says the Cybercab monthly fee should be about $115

On Nov. 14, 2026 Elon Musk wrote that the Cybercab would be available nationwide as Tesla receives regulatory approval, and that the $299 monthly fee is too high and should match Full Self-Driving plus connectivity on other models, about $115 a month. TechCrunch and an archive of the reply to Marques Brownlee quote that post. Musk did not say the listed order price had changed, and the Nov. 13 invitation still showed $299.

Altimeter and Atreides ask Nvidia not to rush a CEO replacement

Altimeter Capital and Atreides Management asked Nvidia's board for a slow CEO search that leaves open a return by Jensen Huang, CNBC reported from a letter Brad Gerstner posted on X on Friday, Nov. 13. The letter says interim CEO Colette Kress can lead in the meantime, that there are more questions than answers, and that it does not demand immediate reinstatement. The firms' June 30 disclosures showed about 9.4 million Nvidia shares at Altimeter, nearly $1.9 billion, and 1.2 million shares at Atreides, nearly $250 million.

AITO launches the updated M8 with Huawei ADS 5

AITO launched the updated M8 on Nov. 13 from 379,800 to 469,800 yuan, in extended-range and battery-electric versions, and Huawei's Richard Yu said it uses an architecture designed for Level 3 driving and carries Qiankun ADS 5. Yu said Huawei will keep leading core technology, product definition, and product design while Seres leads manufacturing, sales, and service, which conflicts with a Sept. 15 notice that Seres would lead product definition and design. He said Seres spent 1.5 billion yuan on 126 upgrades, and neither account is a Level 3 approval.

DeepSeek opens app search, files, and voice on its API

DeepSeek's Nov. 13 API changelog says deepseek-flash now offers web search, deep-thinking search, file input, URL reading, and safety checks on the Messages, Responses, and Chat Completions APIs. Search, Fetch, OCR, speech-to-text, and text-to-speech are also standalone endpoints, and public-facing apps must turn safety checks on. A pricing page read Nov. 15 lists search and web fetch at $0.75 per 1,000 uses, file parsing and OCR at $0.225 per 1,000 pages, text-to-speech at $0.45 per 10,000 characters, and speech-to-text at $0.116 per hour, with speech half off for an undated limited time.

HD Hyundai and Anduril unveil the 600-ton ASV Alpha

HD Hyundai and Anduril held a Nov. 12 launch ceremony at the Ulsan shipyard for a 600-ton autonomous surface vessel that Chosun Biz and an Asia Business Daily on-site report call ASV Alpha. The Korea Herald and Chosun Biz say Anduril COO Matt Grimm and maritime executive Shane Arnott attended, and that the hull pairs HD Hyundai's VAone with Anduril's Lattice. The reports describe sea trials off Ulsan and further testing in the United States early in 2027. No Anduril or HD Hyundai press release, price, or Navy award was opened.

Anthropic makes Opus 5.5 the default on paid Claude plans

Anthropic's Nov. 12, 2026 guide says Claude Opus 5.5 is now the default model in Claude on Pro, Max, Ultra, and Team plans, and it recommends the model for research, analysis, and finished documents, spreadsheets, and decks. The guide lists token prices of $4 and $20 per million for Opus 5.5, $2 and $10 for Sonnet 5.5, $0.10 and $0.50 for Haiku 5.5 on prompts up to 100,000 tokens, and $8 and $40 for Fable 5.2. It says Pro plans and standard seats on Team and seat-based Enterprise plans do not include Fable 5.2 in their limits outside limited-time launch offers.

Microsoft marks the Microsoft 365 outage Service Restored

A Nov. 14 Microsoft Q&A reply and a compiled page for incident MO1550821 both say Microsoft had marked the outage Service Restored, named a preliminary cause in an earlier network capacity upgrade that surfaced after a recent network change, and said restoration was still continuing for remaining customers. Both accounts say Microsoft called it a technical issue rather than a cybersecurity or AI-agent event, with a post-incident report still ahead. The admin center was not opened. The University of Bath said at 10:20 on Nov. 14 that authentication, Teams, Outlook, and Copilot were restored on its campus.

Anduril says Voyager will back the Marines' first light command posts

Anduril said on Nov. 13, 2026 that its Voyager hardware will provide the compute and networking backbone for every system in the U.S. Marine Corps' initial Operational Command Post-Light order. The company says Voyager packs compute, networking, and GPU processing into three suitcase-sized chassis, including VoyagerVM 4.0 and VoyagerGPU 4.0, and that Marines used the system after flying on commercial aircraft during testing. The post does not state the order's value, the number of systems, or the contract vehicle.

SpaceX catches a Starship upper stage for the first time

SpaceX's Flight 15 page says the Starbase tower arms caught the Starship upper stage on Nov. 14, 2026, for the first time, about 22 hours 32 minutes after liftoff, and that all 26 deployed Starlink V3 satellites had made contact and were raising their orbits. Space.com timed the catch at about 4:20 p.m. EST. The page says the booster's boostback ended early because of low fuel mass, the root cause is under investigation, and the flight termination system triggered after the landing burn shut down while the booster hovered over the Gulf; the flight-timeline table still lists a booster catch at 6 minutes 35 seconds, while CNBC says the booster exploded before it hit the water.

TSMC says employee data appeared on an outside website

TSMC's Nov. 12, 2026 Form 6-K says some current and former employee data was posted on an external website without authorization, that the company has found no evidence its own systems were compromised, and that the information matches data stored with a cloud service provider and is unrelated to operations or customer data. The filing says TSMC is notifying affected people and working with the vendor, investigators, and a cybersecurity firm. Central News Agency reported a seller's claim of 46,000 records and 6.9 GB for 3 bitcoin; the filing does not give a count, name the vendor, or confirm those fields.

Oklo says DOE cleared Groves-1 for unrestricted operations

Oklo said on Nov. 11, 2026 that the Department of Energy cleared the final operations hold point for the Groves-1 isotope test reactor in Lockhart, Texas, allowing unrestricted operations and a shift into research, testing, and training. The company said the reactor, which reached low-power criticality in August under the Reactor Pilot Program, will support control-rod tests, criticality benchmarking, and training for later isotope and Aurora projects. No separate DOE release confirming the Nov. 11 clearance was opened, and the announcement does not say Groves-1 has started isotope production.

Anthropic says Fable 5.1 safety scores depend on the setup

Anthropic's Nov. 13, 2026 research post says Claude Fable 5.1's appropriate-response rate on its multi-turn safety evaluations differed by as much as 34 percentage points between the raw API and claude.ai. It says a system-prompt change, including an instruction not to say that self-harm works, lifted the suicide and self-harm appropriate-response rate to 94 percent on claude.ai from 60 percent on the API. Those figures are Anthropic's own measurements, the post does not say the 34-point maximum is that same comparison, and it says multi-turn API confidence intervals can exceed plus or minus 10 percentage points.

/ Archive

Every earlier story, listed by month.