Anthropic, OpenAI, and Google (via Google DeepMind) have been discussing a shared, voluntary AI safety standards body since at least July 2026, according to reporting first attributed to The Information (September 13–14, 2026) and independently corroborated by multiple outlets with on-the-record quotes, plus a primary essay CASRAI read directly. CASRAI’s own NIKOLAI reference vocabulary already has a named record type — an Industry-Wide Recommendation — for the kind of statement a body like this would exist to formalize.
A note on sourcing, up front: CASRAI could not read The Information‘s original article directly — it sits behind that outlet’s paywall, and a search-engine redirect wrapper blocked automated access. What follows instead rests on two things: independent secondary reporting from several separately bylined outlets that agree on the specifics (Yahoo News, RTTNews, BigGo, The AI Insider, Analytics Insight, Unite.AI), and one genuine primary source CASRAI fetched and read directly — Anthropic CEO Dario Amodei’s own essay, published on his personal site. Any claim that traces only to The Information via another outlet’s paraphrase is flagged as reported, not independently confirmed.
The timeline multiple outlets agree on
Reporting converges on a specific sequence. On July 14, 2026, Google DeepMind co-founder and CEO Demis Hassabis publicly proposed a “U.S.-led standards body” for frontier AI modeled in part on the Financial Industry Regulatory Authority (FINRA) — a public-private, self-regulatory structure rather than a new federal agency. Representatives from Anthropic, OpenAI, and Google are reported to have formed a working group around that proposal shortly afterward and met periodically through the summer.
The conversation became public in stages in September. On September 6, OpenAI chief scientist Jakub Pachocki published an essay (“An Alien Mind”) reportedly describing coordinated slowdown among frontier developers as one option on the table. A September 10 WIRED report said OpenAI had asked members of Congress whether an industry-wide coordinated slowdown would raise antitrust concerns. In a September 11 Fortune interview, OpenAI CEO Sam Altman reportedly said he expected a common industry plan to emerge, without detailing private discussions.
Then, on September 12, 2026, Anthropic CEO Dario Amodei published an essay — “We Must Pace the Frontier” — which CASRAI read directly on Amodei’s own site. Amodei writes plainly: “We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.” He lays out three steps: Anthropic unilaterally granting outside evaluators (like METR) employee-like access to its systems; frontier developers in democratic countries coordinating on common safety standards and pacing; and, where feasible, extending that coordination to include authoritarian governments. Per multiple outlets’ reporting, rival leaders responded within about an hour — Elon Musk endorsed the proposal publicly, Altman said OpenAI would commit to its first step, and Hassabis endorsed the direction and tied it back to his July standards-body proposal.
What’s actually confirmed on the record
On September 15, 2026, OpenAI’s global policy chief, Chris Lehane, spoke on the record about the talks — reported independently by both The AI Insider and Analytics Insight in separately written accounts that agree on substance. Lehane confirmed the three companies had been working together on safety “for several weeks” and, on the specific antitrust question, said they do not believe they need a government or antitrust waiver to continue this kind of voluntary coordination. He framed the effort as complementing, not replacing, mandatory federal safeguards, and said OpenAI separately backs a provision in the proposed FRONTIER Act that would require frontier labs to let independent verification organizations assess their safety practices. Because two separately bylined outlets independently reported the same on-record statement from a named OpenAI official, and because CASRAI independently confirmed the underlying essay it responds to, CASRAI treats the existence of the talks, and OpenAI’s “no antitrust waiver” position specifically, as reasonably well-established — even without direct access to The Information‘s original article.
The antitrust question is not one-sided
It would be a mistake to read this as “OpenAI says antitrust law is a non-issue, full stop.” Amodei’s own essay, in the passage CASRAI read directly, takes a more qualified position on the deeper coordination step: “For antitrust reasons, it’s helpful for the US government to mediate or at least enable these discussions — they don’t need to participate, but do need to issue a narrow waiver for certain kinds of safety conversations.” That is a distinct claim from Lehane’s later statement that the current, narrower safety-coordination talks don’t need a waiver. Both can be true: a lab can say today’s talks don’t require special antitrust treatment while its own CEO has separately argued that a more ambitious version — explicit, government-enabled pacing agreements between competitors — would benefit from one.
Other voices push back on the idea itself, not just its legal footing. Cohere CEO Aidan Gomez is reported to have warned that letting the largest AI companies jointly set the rules risks locking out smaller competitors from shaping standards they’ll ultimately have to meet. A White House draft executive order touching similar ground reportedly lost momentum over the summer. None of this confirms or denies that a formal body will actually be created — it establishes that industry-set AI safety standards, coordinated or otherwise, were already contested terrain before this reporting surfaced.
A separate, related story: a lawsuit over the same coordination
This same sequence of events is now also the subject of active litigation, which CASRAI has covered separately: on September 18, 2026, four consumers sued Anthropic, OpenAI, SpaceXAI, and Google in federal court, alleging that the rapid, near-simultaneous endorsements of Amodei’s pacing essay amounted to an unlawful Sherman Act agreement to restrict output, rather than four companies independently reaching the same conclusion. See CASRAI’s coverage of the Sherman Act suit and what the complaint actually alleges for that angle, including its own NIKOLAI crosswalk. This article’s focus is narrower and earlier in the chain: what is and isn’t confirmed about the reported joint standards body talks themselves, separate from the question the lawsuit raises about whether the pacing endorsements crossed a legal line.
Where NIKOLAI fits
CASRAI’s own NIKOLAI project — an independent, unendorsed reference vocabulary for frontier-AI-safety elements, not a standard any lab or regulator has adopted — already has a record type built for exactly this situation. Under NIKOLAI’s Commitments and Governance track (N9), the Industry-Wide Recommendation element is defined as a standard a developer or body states all frontier developers should meet, recorded separately from whether the author itself currently meets that standard. NIKOLAI’s own sourcing for that element already cites Anthropic’s August 2026 Risk Report (“we do not expect to meet our ambitious industry-wide recommendations… in time,” §4.8) and Google DeepMind’s Frontier Safety Framework v3.1, which recommends “a security level for each CCL… the field of frontier AI should apply” (§2.2). Two of the three labs reportedly at this table, in other words, had already independently published the kind of industry-wide recommendation a formal standards body would exist to coordinate, months before this reporting. A body of the kind Hassabis proposed would be a mechanism for turning separate, individually authored recommendations like those into one negotiated set — the gap between “recommendation” and “agreed standard” that NIKOLAI’s element definition is built to keep visible. This is CASRAI’s own editorial mapping, not a claim that Anthropic, OpenAI, or Google DeepMind use or endorse NIKOLAI’s terms.
Key facts, at a glance
- Standards-body proposal: Demis Hassabis (Google DeepMind), publicly proposed July 14, 2026 — a FINRA-style, U.S.-led, public-private body.
- Working-group talks: Anthropic, OpenAI, and Google, reportedly ongoing since shortly after July 14, first reported by The Information, September 13–14, 2026 (paywalled; not directly read by CASRAI).
- Pacing essay: Dario Amodei (Anthropic), “We Must Pace the Frontier,” published September 12, 2026 — read directly by CASRAI.
- On-record confirmation of the talks: OpenAI policy chief Chris Lehane, September 15, 2026 — independently reported by two outlets.
- OpenAI’s stated antitrust position (current talks): no government/antitrust waiver believed necessary.
- Amodei’s stated antitrust position (deeper coordination): a narrow government waiver “for certain kinds of safety conversations” would be helpful.
- Unresolved: whether a formal body is actually created, its governance, and how (or whether) smaller labs get a seat.
Frequently asked questions
Is a joint Anthropic-OpenAI-Google AI safety standards body confirmed to exist?
No. As of this writing it is reported working-group discussion, catalyzed by Demis Hassabis’s July 2026 proposal, not an announced or operating body. OpenAI has confirmed on the record that discussions are happening; no company has announced a finalized structure, governance model, or launch date.
Does OpenAI need an antitrust exemption to do this?
OpenAI’s policy chief says no, for the coordination currently underway. Anthropic CEO Dario Amodei’s own essay takes a more qualified position on a deeper form of coordination, saying a narrow government waiver “for certain kinds of safety conversations” would be helpful. Those are two different claims about two different scopes of coordination.
Is this the same as the lawsuit against these companies?
No, though the two are closely related. A September 18, 2026 consumer lawsuit alleges the rapid public endorsements of Amodei’s pacing essay amounted to an unlawful antitrust agreement — see CASRAI’s separate coverage of that Sherman Act complaint. This article covers the earlier, broader question of the reported standards-body talks themselves.
Is this the same as the NIST-run network of AI safety institutes?
No. That network — see CASRAI’s coverage of the international AI safety institute network and NIST’s Center for AI Standards and Innovation (CAISI) — is government-to-government. The reported Anthropic-OpenAI-Google talks are lab-to-lab, industry-led, and voluntary.
Has this precedent happened before?
Partially. Frontier labs have already formed at least one voluntary cross-lab body — see CASRAI’s guide to the Frontier Model Forum — and multiple labs have made individual, non-identical safety commitments to regulators. What’s newly reported here is a specifically Anthropic-OpenAI-Google conversation about a shared evaluation-and-standards body, distinct from those earlier efforts.
Will CASRAI update this article?
Yes. This article will be revised if CASRAI obtains direct access to The Information‘s original reporting, if a formal announcement is made, or if the reported talks and the pending lawsuit materially converge.
Sources
Original report: The Information, September 13–14, 2026 (paywalled; not directly accessed by CASRAI). Primary source read directly by CASRAI: Dario Amodei, “We Must Pace the Frontier,” darioamodei.com, September 12, 2026. Secondary reporting relied on: Yahoo News, “Anthropic, Google, and OpenAI discussing AI safety standards body,” September 14, 2026; RTTNews, “OpenAI, Anthropic And Google Discuss Joint AI Safety Standards,” September 15, 2026; BigGo, “Anthropic, OpenAI and Google Explore a Joint AI Standards Body as Safety Alarm Grows,” September 14, 2026; The AI Insider, “OpenAI Confirms Weeks of AI Safety Talks With Anthropic and Google DeepMind,” September 16, 2026; Analytics Insight, “OpenAI, Anthropic and Google DeepMind Join Hands on AI Safety Efforts,” September 16, 2026; Unite.AI, Mira Kellan, “Consumers Sue Anthropic, OpenAI, SpaceXAI and Google Over Alleged AI Pact,” September 18, 2026 (timeline corroboration). NIKOLAI element sourcing: Anthropic, Risk Report, August 2026, §4.8; Google DeepMind, Frontier Safety Framework v3.1, §2.2.
