Altman Tells Staff Rivals May Need a Private AI Standards Body Without Washington
Reporting on September 2026 disclosures that Anthropic, OpenAI, and Google have held working group talks since July on a shared AI safety standards body, and that Sam Altman told an internal meeting major labs may need to build one without federal backing.
Classy aggregates publicly available material; we did not conduct a private interview.
What changed
On 14 September 2026, CNN reported that Anthropic, OpenAI, and Google have discussed creating an industry led AI standards body, citing people familiar with the talks. The Information had reported the same working group meetings on 13 September. According to those accounts, executive level teams from the three companies have met regularly since July 2026 to sketch shared testing and audit protocols for frontier models.
The reporting landed days after Anthropic chief executive Dario Amodei published his 12 September essay We Must Pace the Frontier, which calls for embedded third party evaluators with employee level access. Sam Altman publicly agreed with that step. Separately, The Information said Altman told an OpenAI all hands meeting that if Washington does not back a referee, the largest labs may need to construct one themselves.
Google DeepMind chief executive Demis Hassabis had proposed a FINRA style Frontier AI Standards Body in a 14 July 2026 essay, with voluntary pre release review windows that could later become mandatory for U.S. deployment.
Why it matters
Enterprise buyers, insurers, and regulators have been asking for a single, repeatable pre deployment test lane rather than seven incompatible vendor PDFs. A private standards body would not replace law, but it could define what “frontier class” means, what gets tested before release, and whether a model cleared a shared bar.
The CNN account stresses the talks predated Amodei’s public slowdown essay and a former Anthropic researcher’s resignation letter about lab responsibility. That timing matters: the working group is not merely reactive theater. It appears to be an attempt to lock industry rules while White House design work on a federal template remains contested.
Who is affected
Frontier lab safety and policy leads must decide whether to fund a shared body, share models pre release, and accept held out benchmarks they do not control.
Enterprise procurement and risk teams would gain a potential third party pass/fail signal for agent deployments, if the body publishes results with enough detail to inform contracts.
Open source builders are explicitly named in Hassabis’s proposal as board participants, which affects whether “frontier” definitions exclude or include open weights releases.
Policymakers face a fork: bless an industry funded FINRA analogue under federal oversight, or watch labs self certify while capability races continue.
What to do next
If your organization buys or resells frontier models, ask vendors whether they participate in the July working group and what pre release artifacts they would share with an independent body. Treat “we support safety” posts as insufficient until participation terms exist in writing.
What to watch
Whether the three companies publish a charter, funding formula, and benchmark refresh cadence before the next major model drop. Seoul Economic Daily reported on 14 September that teams continued talks for months to build a voluntary framework without government involvement, while Hassabis’s essay assumes federal oversight. The gap between those models is the negotiation to track.
Also watch OpenAI’s promised detail on evaluator access Altman matched on 12 September. A standards body without embedded evaluators would test only finished weights, not training pipelines Amodei emphasized.
Sources
- Primary. CNN Business, Top AI companies have discussed creating their own standards body (14 September 2026). Establishes working group talks among Anthropic, OpenAI, and Google and ties them to Hassabis’s July FINRA style proposal.
- Primary. Demis Hassabis, A Framework for Frontier AI and the Dawning of a New Age (14 July 2026). Defines voluntary 30 day pre release review and path to mandatory U.S. deployment clearance.
- Secondary. Seoul Economic Daily, Rivals Unite: Anthropic, OpenAI, Google Push AI Standards Body (14 September 2026). Adds Altman all hands comment that labs may need to build a body without government support.
- Secondary. InvestingLive, Anthropic, OpenAI and Google held talks on AI safety body, report says (14 September 2026). Summarizes The Information reporting on shared testing protocols and differing views on government role.