three stories about who gets to check AI companies' work -- OpenAI, Anthropic, and Google

1) OpenAI, Anthropic, and Google DeepMind Confirm Weeks of Safety Talks,

On September 16, OpenAI's global policy chief Chris Lehane confirmed to reporters in Washington that OpenAI, Anthropic, and Google DeepMind have spent weeks quietly discussing a shared approach to AI safety -- talks that trace back to a July essay by DeepMind CEO Demis Hassabis proposing a FINRA-style, US-led standards body to test frontier models before release. The catalyst turned public on September 12, when Anthropic CEO Dario Amodei published "We Must Pace the Frontier," urging labs to deliberately slow model releases by a year or two so safety research can catch up -- the same essay President Trump mocked him for by name two days later. Amodei and OpenAI's Sam Altman have now both said their companies will give independent evaluators like METR and Redwood Research unprecedented access to test systems before deployment; Meta and xAI haven't made the same commitment, and Google DeepMind itself, despite proposing the standards body, hasn't committed to embedding evaluators either. The concept, simply: an "independent evaluator" is an outside group with no stake in a lab's release timeline that gets to test a not-yet-released model and report what it finds -- the same relationship an outside auditor has to a company's books, instead of the company grading its own homework. A "standards body" would be a shared industry group, not a government agency, that decides what tests every major model has to pass before shipping.

Why it matters: three fierce competitors are agreeing to let outsiders slow them down together, at the exact moment a sitting president is publicly punishing the one CEO who said the quiet part out loud. Whether the pact means anything depends on whether Meta and xAI eventually join it -- a voluntary safety standard that only half the frontier labs follow doesn't actually raise the floor.

2) A Paperwork Gap Let Banned Nvidia Chips Keep Flowing to China -- Days

An investigation published this week found that Aivres, the US-based subsidiary of blacklisted Chinese server maker Inspur Group, exported at least $5.6 billion in advanced computing hardware to Southeast Asia between April 2024 and February 2026 -- including more than $3 billion in servers built around Nvidia's most advanced Blackwell chips. Because Aivres itself was never separately named on Washington's Entity List, even though parent company Inspur was blacklisted back in 2023, the shipments went through legally while ultimately serving Chinese customers including ByteDance and Alibaba. The finding surfaced days before a planned mid-September US-China AI summit -- the first official AI dialogue between Washington and Beijing since Trump's second term began -- that Treasury Secretary Scott Bessent is expected to lead. The concept, simply: an Entity List ban only restricts the specific legal entities named on it. If a blacklisted parent company has a US-based subsidiary that was never separately listed, that subsidiary can keep buying and re-exporting the same restricted chips completely legally, just one corporate address away from the ban. Why it matters: this is the second time in a year a paperwork gap, not an actual smuggling ring, turns out to be doing more to move restricted chips into China -- and it lands right as both governments are supposed to sit down and negotiate AI chip rules in good faith.

3) Autonomous Coding Startup Factory Triples Its Valuation to $5 Billion in Five

Factory, the startup behind a fleet of autonomous coding agents it calls "Droids," raised $200 million this week at a $5 billion valuation -- its third raise in five months, up from $1.5 billion in April and $4 billion in July. Droids are built to handle full engineering workflows without a human in the loop: triaging an incident, refactoring a module, writing and running their own tests, and moving between those tasks the way a human engineer would context-switch across tools. The round, backed by Blackstone, Sequoia Capital, and Khosla Ventures among others, lands as companies including Coinbase and Block publicly restructure engineering teams around AI-agent productivity gains.

The concept, simply: most AI coding tools still wait for a person to ask a question and hand back a suggestion. An "autonomous coding agent" instead gets assigned a task the way a teammate would, and is trusted to plan the steps, edit across multiple files, run the test suite, and only come back once the job is done or it's stuck. Why it matters: a startup tripling its valuation every couple of months isn't just a funding story -- it's investors pricing in that autonomous engineering agents are close enough to replacing chunks of human engineering work that enterprises are already reorganizing teams around them, before the technology has finished proving itself on the benchmarks the industry uses to measure it.