Following yesterday's four-lab pacing agreement, AI-linked stocks sold off in the first trading session. The Guardian reports that SoftBank fell 13% and South Korea's Kospi fell 3%.
Read source◆ Braid Daily · 2026-09-14
AI's slowdown call hits the market
AI labs called for slower frontier development; the first trading session brought a 13% SoftBank drop and a 3% Kospi decline.
The lead
1Pacing meets institutions
5Altman says pacing needn't wait for Congress
Sam Altman on X
Altman says OpenAI supports a federal framework but doesn't need an antitrust exemption or a new law to pace development. That puts voluntary action ahead of legislation.
Read sourceMicrosoft publishes a 37-page humanist AI code
The Verge
Microsoft published a 37-page humanist AI code of conduct and backed deliberate pacing. It is the first concrete corporate policy document to follow the weekend's statements.
Read sourceWashington has positions, but no shared plan
Axios
Trump rejects a slowdown, House Speaker Mike Johnson prefers a summit, and House Democrats plan to meet Tuesday. Axios says federal consensus is unlikely this year and probably through early 2029.
Read sourceChina rejects the slowdown call
NBC News
China's Foreign Ministry describes the U.S. executives' warnings as fearmongering and a bid to contain China's progress. Chinese officials instead call for openness and inclusivity.
Read sourceUK lawmakers propose independent AI oversight
The Guardian
A cross-party committee of MPs and Lords calls for an independent oversight body and new human-rights legislation. Its concerns include public face scanning and deepfakes.
Read sourceCapital keeps moving
4Anthropic reportedly commits $13.7B to a planned data center
The Information via Techmeme
The Information, citing unnamed sources, reports a six-year compute lease from Anthropic with Rum Group for a planned Georgia data center. Rum Group, which operates Rumble, also hosts Truth Social.
Read sourceAnthropic reportedly posts another profitable quarter
Financial Times via Techmeme
The Financial Times, citing unnamed sources, says Anthropic told investors it was profitable for a second consecutive quarter. Reported gross margins exceeded 80% before partner revenue sharing and training costs.
Read sourceSoftBank secures $11.87B for its OpenAI investment
Bloomberg via Techmeme
Bloomberg, citing unnamed sources, reports an $11.87 billion SoftBank loan facility with commitments from about 20 banks. The two-year facility exceeded SoftBank's $10 billion target.
Read sourceHyperscalers start paying for local consent
The Information via Techmeme
Amazon, Microsoft, Google, and other data-center developers are offering power-cost concessions and community benefits to win local approvals. The permitting dispute is becoming a negotiation over who pays.
Read sourceCapability under deployment
4Anthropic says Claude Code was used for missile guidance
Clash Report
A secondary outlet reports Anthropic's claim that Houthi operators used Claude Code to develop missile-guidance software. The sourcing chain is indirect, so this remains a reported company claim.
Read sourceExperts dispute the evidence behind AI bioweapons warnings
Science
Science surveys expert disagreement over a recent AI-bioweapons report. The dispute adds uncertainty to a week of increasingly specific misuse claims.
Read sourceA preprint serves 64 robots from four GPUs
arXiv
Robion schedules multiple vision-language-action models on shared edge servers. Its authors report serving 64 robots from four GPUs. They met latency objectives for 98% of requests.
Read sourceA small poisoning set triggers targeted robot actions
arXiv
A preprint reports that poisoning 0.31% of training episodes made a targeted robot action fire while nominal task performance stayed intact. The attack transferred across evaluation suites and was demonstrated on a Franka arm.
Read sourceCompanion episode