Today in AI — November 8, 2026

The day's biggest AI story was a governance story. Nvidia's board opened an investigation into whether its chief executive was targeted by Chinese intelligence, hours after he told the company the relationship at the centre of it was real — and the rest of Sunday's news was about who pays: for power, for paper, for the paperwork.

Industry

• Nvidia's board opened an independent investigation into whether CEO Jensen Huang was the target of a foreign intelligence effort, hours after Huang told staff he had been unfaithful to his wife. A committee of independent directors, assisted by outside counsel, will run the inquiry, and the board said it had "no reason to believe that either national security or NVIDIA secrets were compromised" while expressing confidence in Huang's leadership (the board's statement). The memo went further than the New York Times report of October 29: "I was unfaithful to my wife. My actions hurt her and our children. There is no excuse for what I did." The woman is Alice Huang, an executive at Megaspeed, an Nvidia customer already under a Commerce Department investigation over alleged chip smuggling; US officials believe China's Ministry of State Security recruited her, according to the Financial Times. That is what turns a private matter into a national-security one — the CEO closest to this administration's AI deliberations may have been talking to someone reporting to Beijing — and Nvidia says he stays CEO "without interruption" while the review runs.

• Anthropic committed about $3.4 billion to 1.7 gigawatts of gas-fired generators, bought from equipment supplier OnePWR in late September without acquiring the company. The FT reported it is the largest known purchase of power-generation equipment by an AI lab, and the logic is speed and scarcity: buying the machines outright gets a lab past the interconnection queue and past every competitor bidding for the same turbines. It also puts a generation asset on the balance sheet instead of a utility bill, which is a different kind of climate exposure than the one AI companies usually report.

• Apple's new chief executive John Ternus has pressed software head Craig Federighi to move past Siri tweaks, with an autonomous, agentic Siri targeted as early as next year. Bloomberg's Mark Gurman reports the closed-door meeting happened in recent weeks, that the internal comparison set is ChatGPT Dots, Meta Muse and Grok, and that Apple is reworking how it develops iOS, iPadOS and macOS 28 to give engineers more room before features lock. The risk is personnel as much as architecture: Mike Rockwell, credited with rescuing Siri AI, is still seriously considering leaving. Apple owns the devices and still has no computer-use agent to run on them — an odd gap to carry into a year when its rivals ship theirs.

• AI companies have sold or priced more than $300 billion of stock in 2026, with about $100 billion more queued and Anthropic's IPO excluded from the count. That is The Information's tally, and the timing matters: it lands days after the Nasdaq 100 fell 6 percent and roughly $2.3 trillion of market value disappeared in a single session, so the open question is no longer whether the buildout is funded but who is left to buy the paper that funds it. Bloomberg's read on the same trade is that money is already rotating — investors are reviving forgotten Chinese and Indian positions as the US AI trade unwinds.

Policy

• Congressional pressure arrived the same afternoon, with bipartisan lawmakers demanding answers and House panel leaders asking Huang to brief them. CNBC reported the scrutiny is coming from both parties, and the New York Times reported that Marco Rubio had warned officials about the entanglement before it became public, which makes the question who in Washington knew and when rather than what happened in Huang's personal life. Export-control politics sharpen all of it: Nvidia's China revenue runs on licences a hostile Congress can slow down, and a board inquiry that finds nothing still leaves a CEO explaining himself to committees that control his market access.

• The midterm post-mortems found an AI-safety coalition that spent a fraction of its opponent and won most of its races. Anthropic-backed Public First spent more than $63 million and 10 of the 17 congressional candidates it backed won, Bloomberg reported, against roughly $100 million from the industry-aligned Leading the Future. Politico's account of how that happened is the more useful document: a small network that organized in Signal chats turned data-center backlash into a coalition of safety hawks, populists and child-safety advocates, in a cycle where tech-issue campaign ads went from six across 2020 and 2024 to 387, by Reset.Tech's count. The lesson candidates will carry into 2028 is that opposing the buildout is cheaper than defending it.

• Britain's public sector is being buried by AI-generated correspondence, a problem researchers have started calling "agentic flooding." Caseworkers in MPs' offices say roughly half their email is machine-written; the average complaint to a local authority has gone from about two pages to ten times that, with one at 200 pages, according to the body representing 5,600 council lawyers (the Guardian reported). The numbers that make it a policy story rather than an anecdote: employment tribunal claims rose 28 percent year on year in the second quarter to 14,000 single claims, with 70,000 still open at the end of June, and Transport for London took a record 4,769 freedom-of-information requests in 2025-26, with appeals it believes are AI-drafted. Councils are now discussing word limits, templates and fees for requests — a strange place to arrive at, a year into the productivity pitch.

• Environmental groups are using AI to analyse climate data, track wildfires and count endangered wildlife while warning about the footprint of the data centers that make it possible. Reuters found the two positions sitting in the same organizations rather than in opposition. The distinction that keeps getting lost is which machine learning is being discussed: a satellite image classifier and a frontier agent draw on the same grid, and only one of them needs a gigawatt campus.

Tools

• DeepSeek's open-source agent app Harness passed 1.54 million daily active users in the week to November 4, up from 429,000 a month earlier. The company published usage data for the first time, and the growth is steep from a small base — 271,000 at the end of September — per Yicai Global. The number to keep is the token figure: about 92 trillion a day, roughly 60 million per user, which is what an agent that works all day costs to run rather than one that answers questions. Some 35,000 third-party plugins already exist, with an official marketplace and author programme next — the ecosystem question we flagged when the harness's Super-Agent bundle shipped with observation as the default.

What to watch

Monday's open, for a start: no one has priced the Nvidia headline yet. Then whether the board's confidence statement survives a week of committee requests, and whether other labs follow Anthropic into buying generators outright — the first Q4 capex call that lists turbines instead of megawatt contracts will date the moment this stopped being a procurement story.

If a company's biggest national-security risk turns out to be its CEO's personal life, what does a board review actually certify? Write to us via our contact page.