Altman Wants an AI Truce. The Public May Not Get to Read the Terms.

A Senate proposal could protect coordinated safety delays. It would also shield the government notices explaining them from public records requests.

In this storyOpenAIAnthropicNVDA
Vincent JiangVincent JiangSeptember 11, 2026 · 3 min read
Share
Sam Altman speaking at the G20 Innovation Ministerial
Sam Altman at the G20 Innovation Ministerial. The exposure date was not separately established. (Credit: Matt Ramey / AFP via Getty Images, via The Verge, September 9, 2026. Rights clearance pending.)

He left before the payday

Jacob Coxon says he left Anthropic before his equity vested.1

His resignation on X had passed 100 million views by September 9. That figure is a dated observation from the reporting, not a live count.2

Who controls the slowdown

Sam Altman discussed pacing OpenAI's development alongside willing rivals, according to a September 11 staff-meeting report.3

OpenAI also asked Congress whether coordination would be legal, according to September 10 reporting. The stakes include who sets the limits and what the public can learn about them.4

The notices would stay behind closed doors

S. 5105, introduced July 23, offers a possible route. It would protect certain coordinated safety delays from antitrust liability, subject to conditions including advance notice to the Justice Department.5

Section 3(e) would exempt those notices, and information revealing their contents, from public Freedom of Information Act disclosure. It would not prohibit voluntary publication. The bill does not establish that OpenAI has sought this secrecy provision. 5

Billions behind the brake

OpenAI told investors quarterly revenue rose from $5.7 billion in Q1 to $6.7 billion in Q2 2026.6

Separate August 13 reporting put its annualized revenue pace above $40 billion. That extrapolation is not money already earned over a year.7

Column chart of OpenAI quarterly revenue: $5.7 billion in Q1 2026 rising to $6.7 billion in Q2 2026. Company-claimed, unaudited.
Revenue shows the business exposed to new limits. It is not a measure of research speed, profitability, or safety. Company-claimed and unaudited; reported August 18, 2026.

Revenue shows the business exposed to new limits. It cannot establish whether research is slowing, or whether any proposed restriction addresses a real risk.

Secrecy has a case

Nvidia's Jensen Huang framed AI fears as cybersecurity salesmanship.8

Yet OpenAI says it already slowed research after a breach involving its agents. Legal uncertainty can also deter useful collaboration, argues Nicholas Felstead.9,10

The bill requires good faith and an exclusive covered security purpose, and permits government injunctions. Confidentiality can protect vulnerability details; public accountability still needs an answer.5

The terms to watch

Watch the next bill text and any published pacing agreement for specific triggers, an end date, independent review, and a public explanation.

The next test is who gets to inspect the brake before the labs press it.

How this brief was made

01Gathered & sourced182 channels · 1,578 articles

Agents swept 182 channels and ingested 1,578 articles, then de-duplicated and ranked them for signal.

02Verified & cross-validated10 claims · 9 data feeds
03Reviewed & edited2 human editors

2 editors read the draft against the evidence, tuned the framing, and signed off before it shipped.

Confidential tips

Know something about this story? We protect our sources. Reach the editor directly, or read our guide to sharing securely.

Share

Deepdive

AI-generated from this story and its cited sources. Not investment advice.

Reader comments

0 comments

    Sign up

    Get your curated digest

    After email confirmation, you will receive a daily digest of the most relevant news that matter to your portfolio