Afternoon Brief, November 8th 2026
Nvidia's board opened an investigation into Jensen Huang, and Huang told staff he had an affair. Anthropic paid $3.4 billion for gas generators. Claude Code has been saving secrets into files that end up on GitHub. And a researcher shows AI auditors drowning reviewers in false alarms.
Nvidia's board opens an independent investigation, and Jensen Huang tells staff he had an affair
Nvidia's board said on Sunday that it is "initiating an independent investigation" after reports that its founder and chief executive was targeted by a foreign intelligence effort. A committee of independent directors, with outside counsel, will run it. The board says it has "no reason to believe" national security or Nvidia secrets were compromised, that Huang "fully supports the review," and that it has "confidence in Jensen and his leadership team."
The same day Nvidia published a message from Huang to employees. "I owe my family and all of you an apology," he wrote. "I was unfaithful to my wife." The message does not name the woman. TechCrunch and others tie it to Alice Huang, the former Megaspeed executive whom a US intelligence memo, reported on Friday, describes as recruited by China's Ministry of State Security to gather information on him. A company spokesman said "today's news will not slow" Nvidia and that Huang will stay on as chief executive.
Huang is the industry's main advocate in Washington for selling chips to China, which makes this an export-policy question as well as a governance one. The board has named no committee members and given no timetable. NBC News says it has not independently confirmed the New York Times and Financial Times reports behind the memo. SourcesABBB
Anthropic paid about $3.4 billion for 1.7 gigawatts of gas generators, the FT reports
Anthropic bought commitments for 1.7 of heavy-duty gas-fired generators from OnePWR Solutions, a San Antonio equipment supplier, in late September, the Financial Times reported on Sunday, in a deal worth about $3.4 billion. Anthropic first planned to buy power from OnePWR and instead took over its Caterpillar generator orders, with the engineering and construction work needed to install them. It did not buy the company, and no OnePWR staff moved over.
That works out to about $2 million a megawatt for equipment alone. The FT calls it the largest known single purchase of power equipment by an AI lab. It also puts the lab that sells itself on safety into the business of burning gas on site, five days after voters in several states rejected data centers at the ballot box. Neither company commented, and Reuters could not verify the report. SourcesBB
Anthropic tells enterprise customers an outside security firm saw their plan and spend data
Anthropic emailed business customers that a third-party security research company had gained unauthorized access to their contact and account information, according to a notice posted on Saturday by the security researcher Jamieson O'Reilly. For enterprise accounts it lists account and organization IDs, plan tier, usage and spend. Conversations, passwords and payment details were not included. Anthropic says it cut off the firm's access, the firm certified in writing that it deleted the data, and it reported the incident to "the relevant Korean authorities."
The notice does not name the firm or say how many customers were affected, and says Anthropic "cannot confirm whether any particular person's information was viewed." Its description matches the Cathedral incident, in which AI agents reached Anthropic's sales, HR and IT systems through third-party software in under twelve hours in late October. Anthropic has not said publicly that the two are the same. SourcesCC
A judge lets Claude Max buyers pursue Anthropic over its "5x or 20x more usage" claim
Magistrate Judge Alex Tse in the Northern District of California let negligent misrepresentation and breach of contract claims proceed in Kahn v. Anthropic, a over the $100 and $200 Claude Max plans, TNW reported on Saturday from Courthouse News. Anthropic sold them as "5x or 20x more usage" than Pro. The plaintiffs calculate that for one model the weekly limits work out to about 3.5 times and 6 times Pro.
The judge dismissed the claims under California's unfair competition and false advertising laws and a claim over tightening limits mid-billing period, and let one plaintiff amend. It is an early ruling on the pleadings, and it lands on a company that is days from marketing its . SourcesC
AI auditors catch most agent misbehavior, but at 99% detection the best one flags 15% of safe runs
Andy Hall, a researcher at the Anthropic Institute on leave from Stanford, tested five models as auditors of agent logs, writing in a personal capacity. On 's reward-hacking cases, GPT-5 caught nearly 98% at a 5% false-alarm rate, at about an eighth of GPT-6 Astra's cost, and outscored it. Gemini 3.8 Flash caught about half. On , all five caught 71% to 81%. Pushed to 99% detection on a separate set of unsafe actions, the best model, GPT-6 Astra, flagged 15% of safe ones, and Gemini 3.8 Flash flagged 81%.
The arithmetic is the finding. At 100,000 sessions a month, a 3% false-alarm rate produces nearly 3,000 false warnings against about 2,000 real ones, and Hall expects human reviewers to stop reading alerts. Newer, costlier models did not reliably audit better, and Fable 5 and Opus 5 refused to run the evaluation. The reward-hacking set is only 47 cases, so one case moves the rate by about two points. SourcesBA
Amazon has a new home robot group built around Fauna's Sprout humanoid
Amazon has organized a Personal Robotics Group around Fauna Robotics, the New York maker of the 3.5-foot Sprout it bought in March, staffed largely from the team that built Astro, CNBC reported on Sunday from job postings and LinkedIn profiles. A deleted listing described robots working in "dynamic home environments," and an October posting sought a field engineer for on-site repair, which implies hardware in customers' homes. Amazon would not give a price or a date, and its spokesman said a gap remains between the lab and the living room.
Astro, at $1,000-plus, reached a few thousand invited customers and its business version was shut within a year. What changed is the money: PitchBook counts $33.4 billion of venture investment in robotics and physical AI in the first half of 2026, against $28.7 billion for all of 2025. SourcesB
Microchip's chief executive says AI buyers "took over the whole factory," and lead times reach 32 weeks
Microchip Technology told investors on November 5th that AI customers had crowded everyone else out of outside assembly and test capacity. Steve Sanghi said they "came in and took over the whole factory." Standard-product have stretched to six to 32 weeks, from four to eight, and moving capacity in-house takes about six months. Shares fell 10% on Friday after a 14% drop on Thursday, and Truist, Raymond James, Citi and JPMorgan all cut their targets.
The quarter itself was strong, with sales of $1.622 billion, up 42% on the year, and data center revenue above $250 million. Sanghi said "some deceleration of the AI demand actually would be quite positive." It is the clearest statement yet that the AI build is taking supply from the rest of the chip industry, from a company that sells into both. SourcesC
Google takes over Crusoe's troubled Wyoming campus and says it will spend $17 billion there
Google announced on Thursday a $17 billion investment in its first Wyoming data center, in Laramie County near Cheyenne, powered by a 702-megawatt natural gas development built by Bloom Energy and contracted through AEP, which Google calls the largest of its kind. Most of the emissions will be captured and stored at Tallgrass's sequestration hub, and the site will be air-cooled to spare local water.
On Friday Crusoe confirmed it had sold Google the unfinished project, which it announced in 2025 with Tallgrass at a planned 10 gigawatts. Google had tried and failed to sign on as a customer and reportedly raised concerns about cost and schedule. The price was not disclosed. SourcesAB
Ramp's card data shows business token spending grew 1.5% across August and September
Business spending on AI grew 1.5% across August and September, after 42% in July alone, according to Ramp's AI Index, which tracks corporate card and bill payments. It rose 8.9% from September 23rd to October 23rd. The median firm spends $12 to $15 per employee a month. The top 1% spend just under $10,000, below their summer peak.
Price cuts explain part of it. OpenAI halved prices for GPT-6 Sol and Luna in September, and Anthropic says in its filing that its own cuts on Fable 5.1 and Opus 5.5 took about $600 million off revenue growth. Lower prices with flat spend means more tokens at less revenue, which is the number IPO buyers are asking Anthropic for. SourcesBA
The Justice Department creates a "Super Intelligence Task Force"
The Justice Department launched a Super Intelligence Task Force on Thursday to prosecute misuse of AI by foreign state actors, criminal organizations, and companies or individuals committing fraud, including schemes against taxpayers, infrastructure and national security. Associate Attorney General Stanley Woodward and Associate Deputy Attorney General Stephen Weiss lead it, and it will work with Jay Clayton as chair of what the release calls the SI Force.
Attorney General Todd Blanche said "SI development will continue to be powered by American innovators." The release also commits the department to promoting innovation and encouraging internal controls at companies, so the same unit that prosecutes misuse also has a mandate to help the industry. SourcesA
A system built on an unreleased Anthropic model proves an 18-year-old quantum computing conjecture
Physical Superintelligence said on Friday that its internal research system, running instances of a non-public Anthropic model, proved the Aaronson-Ambainis conjecture, open since 2008. It implies that quantum computers cannot be exponentially faster than classical ones on random inputs except where the input has rare structure. The system had a complete proof on day three of the run, extending an approach by Shachar Lovett and Jiapeng Zhang with a new inequality it proved itself.
The proof checks in , the computer , and a Stanford group whose August paper settled a special case has read it and sent comments. A human researcher, David Turturean, takes responsibility for the paper's correctness. It is a second result in a week on hard mathematics from AI systems, after DeepMind's agents found two weaknesses in 1024-bit . SourcesA
OpenAI's Images 2.6 makes animated GIFs in ChatGPT and runs about 50% faster
OpenAI released Images 2.6 on Wednesday, adding animated GIFs to ChatGPT and Codex from a text prompt or a photo, with frame-by-frame editing and speed control, on every plan. OpenAI says generation is about 50% faster, and that ChatGPT and the API now produce more than 4 billion images a week, up from 3 billion in September. Developers get two API models, GPT-Image-2.6 Flare for speed and Sunburst for quality.
Motion is back in ChatGPT eight months after OpenAI shut the Sora video app. The GIF feature is rolling out gradually, and OpenAI has not published per-plan limits. SourcesBB
Seven Nova Scotia towns rally against Anthropic's planned data center
Protesters in Halifax, Kentville, Port Hawkesbury, Truro, Bridgewater, Westville and Sydney rallied on Saturday for a on large AI data centers until the province sets rules on water, power and environmental review. Halifax drew several hundred to nearly 1,000, by varying reports. The target is EverWind Fuels' planned campus at Point Tupper on Cape Breton, reported at $10 billion, with Anthropic as the proposed customer: 95 megawatts in 2027 and 325 more in mid-2028, drawing on the provincial grid until EverWind's approved but unbuilt wind farms are running.
Premier Tim Houston's office said the project proceeds only if it meets provincial and federal requirements. The backlash that sank in the US on Tuesday now has a Canadian chapter, aimed at a named AI lab. SourcesC
Google's dealmaker Don Harrison leaves SpaceX's board
SpaceX said on Friday that Don Harrison, Google's president of global partnerships and corporate development, has resigned from its board, in a filing that says the decision "was not due to any disagreement." Alphabet is SpaceX's largest institutional investor, with a stake worth $74 billion, and Google has agreed to pay SpaceX's AI unit up to $920 million a month for compute near Memphis. Google said it remains "deeply supportive" of SpaceX. No reason was given, and the seat leaves Google without a director at a company that is now both a supplier and a competitor in AI. SourcesB
Dropbox pays $350 million for Tigris Data to sell storage to AI agents
Dropbox agreed on Thursday to buy Tigris Data, a storage startup founded by former Uber infrastructure engineers, for $350 million, apparently its largest acquisition. Tigris becomes the base of DBX Agent Cloud, S3-compatible object storage for developers building models and agents, with file system features for agents planned. Dropbox expects the deal to close this quarter and says it has no plans to cut Tigris staff. Tigris raised a $25 million in 2025. SourcesB
Sources
- A Statement from NVIDIA Board of Directors
- B Nvidia will investigate CEO Jensen Huang's relationship with possible Chinese spy (TechCrunch)
- B Nvidia opens investigation after reports CEO Jensen Huang was targeted by Chinese intelligence (NBC News)
- B Nvidia CEO Jensen Huang admits to having an affair with suspected Chinese spy (Tom's Hardware)
- B Anthropic secures $3.4 billion generator deal for AI data centers, FT reports (CNBC)
- B Anthropic buys power for AI data centres in $3.4bn deal, FT says (Bloomberg)
- B Claude Code's "Don't Ask Again" file (The Hacker News)
- C Anthropic emails enterprise customers that an outside security firm may have viewed their contact, plan and spend data (MLQ.ai)
- C Jamieson O'Reilly on X
- C Judge lets Anthropic face Claude Max usage-claims class action (TNW)
- B AI auditors catch most agent misbehavior but flood reviewers with false alarms (The Decoder)
- A We're using AI to audit AI (Andy Hall, Free Systems)
- B Amazon has a new home robot team (CNBC)
- B Apple working on agentic Siri next year, report (9to5Mac)
- C The chip shortage is back and Microchip says AI customers are crowding everyone out (24/7 Wall St.)
- A Our first data center in Wyoming (Google)
- B Crusoe sells troubled Wyoming data center project to Google (TechCrunch)
- B Ramp data shows AI spending stalling just as Anthropic nears its IPO (TechCrunch)
- A Ramp AI Index
- A Department of Justice launches Super Intelligence Task Force
- B Meta pulls down controversial terms of service (Nieman Lab)
- A Aaronson-Ambainis (Physical Superintelligence)
- B ChatGPT can now make animated GIFs with OpenAI's Images 2.6 (TNW)
- B OpenAI ChatGPT animated GIFs, Images 2.6 (The Verge)
- C Nova Scotia: seven rallies demand AI data centre moratorium (Tech Insider)
- B SpaceX says Google exec Don Harrison has resigned from its board (CNBC)
- B Dropbox acquires Tigris Data for $350M (TechCrunch)