Three AI labs, one proposed standards body: what OpenAI confirmed on 15 September 2026, what Anthropic and Google DeepMind have not, and where the FINRA analogy comes from

When competitors say they are cooperating on safety, the useful questions are the boring ones: who said it, to whom, in what words, and who has not said it. On 15 September 2026 the head of global affairs at OpenAI answered the first three. The other two companies, as of the reports we read, have not answered the fourth. This article separates the confirmed statements from the reporting built around them, and traces the FINRA comparison back to the document where it started.

Bloomberg reported, and Reuters carried the report, that Chris Lehane, OpenAI's chief global affairs officer, told reporters at a briefing in Washington that the company's engagement with Anthropic and Google DeepMind on safety "had been under way for several weeks". He said the three companies did not need an antitrust waiver to coordinate on the subject. The sentence quoted across the coverage is his: "It's better to try to work together to prioritize safety." A Korean broadcaster's account of the same briefing adds that he cited aviation as a precedent for competitors cooperating on safety, saying "There are many instances where companies have tried to help each other solve safety issues."

Reuters said it could not independently verify the Bloomberg report and that none of the three companies immediately responded to its requests for comment. That is the state of the record: one company's senior official, on the record, describing talks; two companies silent.

Lehane's remarks did not come from nowhere. Six days earlier, on 9 September 2026, he published a policy note on OpenAI's website. Its operative sentence: "We will work with other frontier labs to advance frontier AI standards, building a voluntary effort now, with or without government support." The note lists what such standards should cover, "measuring capabilities, managing risk, preserving human control, and determining when and how development should slow or stop", and it draws a line that matters for the antitrust question: "Any industry-led standards would complement—not replace—mandatory federal safeguards and democratic oversight." The note also argues for independent technical assessments to replace what it calls a "fragmented system of private governance".

The comparison to the Financial Industry Regulatory Authority, the self-regulatory body that oversees US broker-dealers under the Securities and Exchange Commission, is not the journalists' invention. It comes from an essay Demis Hassabis of Google DeepMind posted on 14 July 2026, titled "A Framework for Frontier AI and the Dawning of a New Age". TechCrunch's report of the essay quotes the core proposal: a "standards body" modelled after FINRA "which could test frontier models and develop best practices for their release".

Hassabis was specific about the sequence. "Initially, Frontier Labs would voluntarily share models with the Standards Body for review up to 30 days before release. Once the assessment protocol is shown to be effective and robust, formalisation could quickly follow, meaning that Frontier Models would be required to pass it to be deployed in the US market." The body would be "backed by the U.S. government but funded by the AI industry and operated independently", staffed by "open source representatives and technical experts from within the industry". On money, he wrote that funding "would need to be substantial and likely mostly come from industry, in order to attract world-class technical talent and provide the necessary compute resources for large-scale testing".

The FINRA model is worth understanding precisely because it is not simple self-regulation. FINRA writes and enforces rules for its member firms, and membership is mandatory for US broker-dealers; but it does so under SEC oversight, and the SEC can approve or reject its rules. Transposed to AI, the proposal is therefore a two-layer structure: an industry-funded technical body doing the testing, and a federal authority above it. Whether that second layer exists yet is the open question.

The coverage also records dissent. Aidan Gomez, chief executive of the Canadian AI company Cohere, is quoted by AFP criticising the effort: "The dispute is over who writes them, who gets to participate and whose interests the rules are protecting." And there is a difference in emphasis between the labs themselves: Anthropic's chief executive has argued publicly for independent auditors and has suggested a narrow antitrust waiver might be needed for certain safety conversations, while Lehane said on 15 September that OpenAI does not think one is necessary. Both can be true at once; they describe different depths of cooperation.

On the political side, Bloomberg's assessment, relayed in the coverage, was that the chance of AI legislation passing this year is slim, with the House not reconvening until after the November midterms. The White House position, as reported the same day, is that existing tools are sufficient to police AI safety.

There is no signed agreement in the public record. There is no named body, no charter, no membership list, no test protocol. What exists, as of 15 September 2026, is a published proposal from one lab's chief executive in July, a published pledge from a second lab in September, and a spoken confirmation from that second lab that talks with the other two are under way. Everything else is inference.

For our own purposes the case is a useful control sample. We track how AI systems document their own work in our Insights series, including last week's piece on Anthropic's threat intelligence report, and the standard we hold ourselves to is the one applied here: a claim is a claim until a document backs it. Readers who want to know how our research desk is organised can start on the team page; the markets we watch, with their sources, are on Market Observation.

• OpenAI, Chris Lehane, policy note on frontier AI standards (openai.com/index/ai-policy-window), 9 September 2026: https://openai.com/index/ai-policy-window/
• Bloomberg, "OpenAI Says It's Working With Anthropic, Google on AI Safety", 15 September 2026: https://www.bloomberg.com/news/articles/2026-09-15/openai-says-it-s-working-with-anthropic-google-on-ai-safety
• The Economic Times (carrying Bloomberg/Reuters), "OpenAI is working with Anthropic, Google on AI safety: Bloomberg News", 15 September 2026: https://m.economictimes.com/tech/technology/openai-is-working-with-anthropic-google-on-ai-safety-bloomberg-news/articleshow/134267292.cms
• AFP via Dawn, "OpenAI, Anthropic and Google are working to create an AI standards body", 15 September 2026: https://www.dawn.com/news/2030165/openai-anthropic-and-google-are-working-to-create-an-ai-standards-body
• SBS News (English), "OpenAI Says It Is Cooperating with Anthropic and Google on AI Safety, No Antitrust Exemption Needed", 16 September 2026: https://news.sbs.co.kr/english/article.do?news_id=N1008754951
• TechCrunch, "DeepMind CEO calls for an independent standards body to regulate frontier AI", 14 July 2026: https://techcrunch.com/2026/07/14/deepmind-ceo-calls-for-an-independent-standards-body-to-regulate-frontier-ai/
• The Daily Record, "Trump dismisses alarm, says US has tools to police AI safety", 15 September 2026: https://thedailyrecord.com/2026/09/15/trump-us-tools-police-ai-safety-concerns/

Educational content. Not investment advice. Data Analytic Investments Kft. is an educational and market-information provider; it is not a crypto-asset service provider and does not offer investment recommendations.