The Three Largest Labs Are Discussing a Self-Run Pre-Release Testing Body, and Cohere's CEO Calls It a Cartel by Another Name

OpenAI's policy chief confirmed on September 18 that the three largest labs have been working for weeks on a FINRA-style body to test models before release. Anthropic, OpenAI and Google DeepMind have been meeting in private working groups since July to explore this self-regulatory body for safety auditing and pre-release testing of high-capability models; the discussions cover third-party evaluations before release, formal safety checks, and unified risk protocols for cyber attacks, bioweapons risks and deceptive behavior. The starting point was a July essay by Google DeepMind founder Demis Hassabis proposing a standards body modeled on the US Financial Industry Regulatory Authority — industry-financed, government-monitored, staffed by independent experts, running evaluations roughly 30 days before release. There is currently no formal charter and no finalized governance setup. The sensitive point is antitrust: Amodei has pushed for aligned benchmarks together with antitrust protection for safety-related coordination, while three direct competitors agreeing to hold back technology can read as restraint of competition to a regulator. Cohere co-founder and CEO Aidan Gomez wrote on his company's site on September 13 that an industry-run standards body of this kind amounts to "a cartel by any other name," arguing the dispute is not over whether there should be rules but over who writes them, who gets to participate, and whose interests they protect.

This Is the Other Half of Yesterday's Story

On September 18 this site covered AEF-1: a third-party evaluation standard that had existed for nine months, which three labs walked over to and cosigned this week. Today's story is the other half of the same thing — **not merely adopting someone else's standard, but running an institution of your own.** The difference between those two is exactly what is being fought over. When you adopt an external standard, the rules' author sits outside. When you run the institution, the rules' author is the handful of companies the rules bind. Gomez's line puts it most clearly: the dispute is not whether there should be rules, but who writes them, who participates, and whose interests they protect. Yesterday we described AEF-1's structural problem as the evaluator's revenue coming from the evaluated. Today's story pushes that one layer further: if the body is industry-financed and founded by the top three, then **even the author of the rules is one of the parties the rules govern.** Hassabis's original proposal included government monitoring, while the current private working groups have no formal charter — the distance between those two points is the thing most worth watching.

The Antitrust Request Says More About This Than Any Criticism Does

The most informative item is not an objection but a request from the proposing side: alongside pushing for aligned benchmarks, Amodei has sought antitrust protection for safety-related coordination. That amounts to conceding the coordination has competition-restricting effects — otherwise no exemption would be needed. It is exactly what Gomez seized on, saying a CEO asking for antitrust law to be relaxed for them should raise a red flag. This is not wordplay. Three direct competitors jointly agreeing on which capabilities may not ship and how long to wait before shipping is, in effect, jointly restricting supply. If the mechanism genuinely reduces risk it deserves discussion — but it needs an adjudicator not defined by those three, because otherwise "safety" and "raising the barrier against later entrants" are indistinguishable from outside.

One of Gomez's Alternatives Already Has Something to Compare Against

Gomez proposes four pillars: an internationally published evidence-based risk framework; mandatory developer transparency through model cards and incident reporting; independent testing scoped only to genuinely dangerous capabilities; and independent assurance mechanisms modeled on financial, aviation and nuclear regulation. The second one has a reference point right now. On September 18 this site covered the misalignment disclosure framework OpenAI had just delivered — with six- and twelve-business-day deadlines and six of its own cases attached. That is the **voluntary version done by a single company**, while Gomez wants a mandatory one. Put side by side, the substance of the disagreement gets concrete: not whether to disclose, but voluntary versus mandatory, and who defines what has to be disclosed. Three boundaries to state: there is no formal charter and no finalized governance arrangement; the three confirm talks but have published no timeline; and Gomez's criticism comes from a direct competitor — which does not weaken his argument, but a reader should know his position, since a vendor outside those three opposing rules written by those three also has a commercial motive. Separately, Altman has voiced support for industry-wide collaboration, saying it is good for the industry to say it wants to come together. As it stands, this is still far from an institution that could actually stop a release: no charter, no adjudicating authority, no consequences for departing from it.

via: CNN Business, The Globe and Mail on Gomez's criticism, Lawfare, "Designing a FINRA for Frontier AI", Free Malaysia Today