Google, OpenAI and Anthropic Are Writing Their Own Rulebook — Inside the Frontier AI Standards Agency

TL;DR

Google, OpenAI and Anthropic are reportedly moving ahead with a voluntary frontier-AI standards body — tentatively the Frontier AI Standards Agency — targeting launch in late 2026 or early 2027, with no government oversight.
The three labs have approached Sriram Krishnan, the White House’s senior AI policy adviser until June 2026, to run it. His stated position in office: there would be no FDA for AI.
The design borrows FINRA’s template but strips out the two mechanisms that give FINRA teeth — SEC supervision, and the power to fine, suspend or expel members.
Cohere CEO Aidan Gomez called it “a cartel by any other name.” The Frontier Model Forum, the 2023 predecessor with largely the same membership and a $10M fund, has never blocked a single release.

Introduction

The three companies that build the most capable models in the world have spent 2026 arguing in public that AI safety needs shared standards. On September 24, the picture of what those standards will actually look like sharpened: a body run by the companies themselves, with the piece that would have given it legal force removed.

The plan — provisionally the Frontier AI Standards Agency, also reported as the Standards Authority for Frontier AI — is being assembled by Google, OpenAI and Anthropic, with a target launch in late 2026 or early 2027. To lead it, the three labs approached Sriram Krishnan, the Trump White House’s senior AI policy adviser until June 2026, after roughly 18 months in the role. (Source : The Information — Google, OpenAI, Anthropic AI Safety Group Takes Shape)

Krishnan’s position while in office was explicit: no new licensing regime, no central regulator, and no government body that would, in his framing, put sand in the gears of AI progress. He compressed it into a line reported in July 2026 — there will be no FDA for AI, a reference to the agency that must approve drugs before they reach market. (Source : TechCrunch — DeepMind CEO calls for an independent standards body to regulate frontier AI)

The remit: four pillars, all voluntary

The agenda under discussion, per summaries of the report, has four components: shared pre-release evaluation protocols, third-party safety testing before a model ships, standardized incident-reporting rules, and qualification standards for the auditors who would conduct those tests. (Source : AI Weekly — Google, OpenAI, Anthropic Court Sriram Krishnan for AI Safety Body)

That is a coherent, technically serious package. It is also the exact list of functions that would need to exist in a real regulator. The difference is the funding, the staffing and the enforcement — none of which have been published. No charter, no budget, no member list, and no definition of what power the body would hold to stop a launch has surfaced. Google and Anthropic have not publicly confirmed the coordination at all.

What changed between July and September. The idea did not start as pure self-regulation. On July 14, Google DeepMind CEO Demis Hassabis proposed a FINRA-style body — explicitly modelled on the private organization that supervises US broker-dealers under SEC oversight — built as a public-private partnership with a federal backstop, a board including independent technical experts and open-source representation, and voluntary submission of models up to 30 days before release. Hassabis’s version included a path to making approval mandatory for selling models in the US market once the method proved out. (Source : TechCrunch — DeepMind CEO calls for an independent standards body)

The version now advancing drops precisely that federal piece. The three labs have been meeting since July; The Information reported the talks on September 13, and an OpenAI spokesperson confirmed them to CNBC two days later. (Source : CNBC — OpenAI, Google and Anthropic discuss AI safety standards) By September 24, the design being described was self-run, self-funded and voluntary.

The timing is hard to read as coincidence. On September 22–23, Sam Altman and Dario Amodei both addressed the UN Security Council calling for common standards between governments and industry — and the US delegation rejected any global governance structure. Two days later, the labs’ own private alternative surfaced. (Source : FomoEra — Google, OpenAI y Anthropic alistan su propio vigilante de IA)

The evaluation bottleneck is the real payload

Strip away the governance debate and there is a concrete, already-measurable effect on how fast models ship.

Standardized pre-release evaluation means longer access windows for outside testers — and access windows are currently very short. METR and Redwood Research spent six days on-site evaluating an OpenAI agent deployment, receiving the full dataset in their final two days. Apollo Research had three days with GPT-6 Astra and only two with chain-of-thought access. (Source : METR — Notes on frontier AI safety regulations)

Formalize that pipeline and the release cadence changes shape. At the G20 Innovation Ministerial, Altman described next-generation models as “sobering for everybody,” paired with an acknowledgment that OpenAI is deliberately throttling what it ships. Enterprise teams that planned 2027 roadmaps around a steady stream of frontier releases should treat cadence, not capability, as the variable that moved. This dynamic connects directly to the development-brake debate Altman and Amodei opened earlier in September — the brake is no longer hypothetical.

The precedent is not encouraging

The Frontier Model Forum already exists. Anthropic, Google, Microsoft and OpenAI formed it in July 2023 with a $10M safety fund and a mandate to advance responsible AI development. In three years it has published best practices and, by every available account, never stopped a release, required an evaluation, or enforced a standard. (Source : ByteIota — Google, OpenAI, and Anthropic Are Building Their Own Regulator)

The structural objection is the one Aidan Gomez, CEO of Cohere, put bluntly: a cartel by any other name. The argument is not that the standards would be bad — it is that a body funded and staffed by the three largest labs can set requirements they clear easily while raising the cost of entry for smaller labs and open-weight developers. It is the same objection that applies to industry self-policing in every domain where the audited pay the auditor, and it is why the release of a charter — who funds it, who joins beyond the trio, whether its tests can delay a launch — is the only verifiable signal worth waiting for.

FAQ

Has the agency launched? No. It has no charter, budget, member list or published enforcement power. The target is late 2026 or early 2027.

Has Sriram Krishnan accepted? No confirmation from Krishnan or any of the three labs. He left the White House in June 2026.

Does it have power to block a model release? Nothing published suggests it does. FINRA can suspend firms, impose fines and bar members from the market; the proposed body inherits none of that.

Why does this matter if it is voluntary? Because the behavior change is already underway. Longer pre-release evaluation windows slow releases even when participation is nominally optional.

What should developers watch? Whether the body gets enforcement power. Until then, treat it as a signal about slower release cadence, not a guarantee of safer models.

Introduction

The remit: four pillars, all voluntary

The evaluation bottleneck is the real payload

The precedent is not encouraging

FAQ

Further Reading