Dario Amodei asked for antitrust cover on Saturday. The Information reports the coordination it would cover has been running since July.
What Happened
The Information reported on Sunday evening that Anthropic, OpenAI, and Google have held working-group meetings since July 2026 on a shared, industry-led safety standards body covering model testing and pre-release audit. The talks predate Amodei’s Saturday essay by roughly two months, continued through last week, and are characterised in the reporting as an effort to build a private-sector-led organisation after the administration’s own discussions failed to progress.
Two government processes are easily confused and need to be kept separate. The first: on 3–4 August, previewed by Bloomberg, CNN, and CNBC, the White House convened AI companies — Anthropic, OpenAI, Google, and Meta among them — to review a voluntary model-testing framework. That framework was reported finalised. It is not the thing that stalled. What stalled is a separate draft executive order to establish a self-regulatory AI organisation, reported to build on a FINRA-style proposal and held up by internal opposition from David Sacks, whose objection — that it would function as a “DMV for AI” — is on the record. Sacks is not against an industry body; on 21 August he proposed one modelled on the Motion Picture Association, administering a voluntary rating system, and described Elon Musk as supportive.
One thing has not happened, and it matters more than anything that has: nothing has been formed. No entity exists. No company has confirmed a launched body. No participant list, charter, governance structure, funding model, or timetable is public. What exists is a set of working-group meetings, a published essay requesting legal accommodation, and a reported internal position at one company that the body will have to be built without government support.
The key insight: Read on Saturday, Amodei’s waiver request looked like an opening bid for a hypothetical future arrangement. Read now — with the July start date in view — it looks like an attempt to put a legal floor under something that had already been happening for two months. The sequence changes the meaning of every public statement made in between.
The Structural Read
The Permission Layer framework — which holds that governance infrastructure controls which AI capabilities can be deployed, by whom, and under what conditions — normally runs in one direction: public process sets the boundary, industry operates inside it. What this story describes is the mechanism that activates when the public process fails to materialise.
When an industry faces a genuine collective-action problem and the public mechanism for solving it stalls, the coordination does not stop. It privatises. That is not concealment — it is the underlying problem refusing to disappear just because the process has. The companies involved face real technical questions about model evaluation that benefit from shared methodology, and those questions existed before the executive order stalled and continued to exist after it did.
But the migration from public forum to private working group is precisely what creates legal exposure. A standards process convened by government carries public legitimacy and, ordinarily, some measure of antitrust protection. The identical technical conversation held privately among the three largest suppliers in a market carries neither. That asymmetry is the structural core of this story. It also reframes Amodei’s request not as abstract policy advocacy but as a practical legal obstacle the participants had, by July, already encountered in practice.
Permission Layer — Coordination Privatises When Public Processes Stall
Safe harbour is what government supplies; without it, the legal form of coordination changes the risk profile entirely
A government-convened standards body gives antitrust cover. A private working group among three companies with dominant market positions does not — even if the technical content of the conversations is identical. The waiver Amodei requested is not decorative. It is the difference between an industry institution and a legal exposure.
Three Statements That Read Differently Now
The sequence makes three public positions newly legible — and one of them turns out to have been exactly accurate.
Sam Altman — Fortune interview
“I’m not going to pre-announce private discussions that I think should be at some point shared as a group.”
Widely read at the time as a polite deflection, that sentence was a precise description of fact. The discussions existed. They were private. There was an expectation that they would eventually be announced jointly. It was not a non-answer; it was an accurate answer that withheld the specifics it described.
Dario Amodei — published essay, 13 September 2026
A narrow antitrust waiver so that rival laboratories could coordinate on safety.
This stops looking like abstract policy advocacy and starts looking like a legal obstacle the participants had already encountered in practice. The waiver request is the tell: you do not ask for a legal floor under a hypothetical arrangement. You ask for one when the arrangement is real.
David Sacks — public statement
“Pretending antitrust law has to be suspended so you can form a cartel.”
Sacks’s objection reads as continuity rather than reaction. He is reported to have held an equivalent position internally since at least August, when he opposed the draft executive order. Nothing establishes that he knew of the labs’ working groups specifically — that connection should not be asserted — but his public framing is consistent with a position he had already staked. He was not reacting to a Saturday essay. He was restating, in public, a view he had already expressed in private process.
A fairness point that matters: discussing technical standards is ordinary and lawful. Standard-setting organisations operate in every major industry and are usually a public good. These companies also asked publicly, in a published essay, for a legal accommodation — which is the opposite of concealment. The antitrust point concerns legal form and exposure, not wrongdoing. Nothing here asserts or implies that anyone has acted unlawfully.
Three Implications
IMPLICATION 1 — The Strategic Turn Is Proceeding Without a Counterparty
Altman is reported to have told his all-hands that the major labs will have to build the body themselves, without government support. Every public design proposed this month assumed a public partner. Amodei’s waiver required one. Google DeepMind’s FINRA-style proposal explicitly described a body backed by the US government, funded by industry, operated independently. Proceeding without that partner produces something structurally different: an industry association with testing functions, no statutory authority, no safe harbour, and no answer to the question Cohere’s Aidan Gomez raised about who gets to write the rules. It may be better than nothing. It is not the thing anyone described.
IMPLICATION 2 — What Remains Unknown Is Most of the Important Part
It is not public whether any body would test models before release, what authority it would hold over a member that failed an assessment, whether membership would be open beyond the three current participants, or how it would relate to the embedded-evaluator commitments Anthropic and OpenAI made last weekend. Those are not implementation details. They are the whole question of whether this becomes a safety institution or a trade association. Note also which companies are not named in the reporting: Meta, SpaceXAI, Microsoft as a first-party model developer, and every laboratory outside the United States. Not named is not the same as excluded — membership may well be an open question — but a three-company working group is the strongest available evidence for taking Cohere’s concern about incumbent-designed standards seriously.
IMPLICATION 3 — Legitimacy Is the Scarce Resource, and Government Was Always Supplying It
Every design — Amodei’s waiver, the FINRA model, the embedded-evaluator commitments — treated government not as a regulator to be tolerated but as the legitimising element that makes an industry body more than a club. Without a statutory anchor, the body has no answer to the question every excluded party will ask: why are these three companies writing the rules? The Permission Layer framework maps this precisely — what stalled was not just a process but the mechanism by which the industry was going to borrow public authority. That authority cannot simply be self-issued.
