AI companies are planning a kind of FINRA for frontier AI – without a constitution, leadership, or formal organization

AI companies are planning a kind of FINRA for frontier AI – without a constitution, leadership, or formal organization

Google, OpenAI and Anthropic are reported to have agreed to establish a joint organisation for AI safety standards, to be called the Standards Authority for Frontier AI.

Google, OpenAI and Anthropic are reported to have agreed to establish a joint organisation for AI safety standards, to be called the Standards Authority for Frontier AI. But as of mid-September 2026, there is no formal organisation, no leader and no charter – only plans described by anonymous sources to The Information.

The core of the news

According to reporting on 24–25 September 2026, primarily at The Information and republished by BankInfoSecurity and CXOToday, the three leading frontier AI labs have agreed to found an independent, industry-led organisation for AI safety standards. The body could launch by the end of 2026 or in early 2027, according to The Information's anonymous sources (BankInfoSecurity).

It is important to be precise about what this means: the entire core of the story rests on anonymous sources to a single outlet, relayed second-hand by others. None of the companies has itself confirmed that the Standards Authority for Frontier AI exists as a named initiative, and no documents – charter, registration or bylaws – have been published. Crypto Briefing puts the status bluntly: as of mid-September 2026, there is no formal organisation, no named leadership and no published charter (Crypto Briefing).

What is somewhat better documented is that coordination is happening. According to Crypto Briefing, Google DeepMind, OpenAI and Anthropic have had working groups running since July 2026, with a target launch in 2026 or 2027 (Crypto Briefing).

What OpenAI's confirmation actually means

The most concrete confirmation does not concern the body itself, but the coordination around it. OpenAI's global policy director Chris Lehane confirmed on 15 September that coordination among the three companies had been a strong focus, and rejected the idea that the collaboration would require exemptions from competition regulations. Lehane characterised the effort as risk management (Crypto Briefing).

This is a company statement, not independent documentation of an established organisation. But it points to a real legal and political question: when direct competitors coordinate standards for their own industry, the question arises whether this is safety work or covert antitrust activity. Lehane's answer – that no exemptions are needed – is therefore a position in a debate that would likely follow the body if it is realised.

Where the idea comes from – two conflicting stories

According to BankInfoSecurity, the proposal originated in a 14 July essay written by Google DeepMind co-founder Demis Hassabis, in which he launched the idea of a self-regulating organisation modelled on FINRA – the Financial Industry Regulatory Authority, the self-regulatory body for the US securities market – with funding from the industry itself (BankInfoSecurity).

Crypto Briefing, however, instead points to Dario Amodei's 12 September essay as the intellectual foundation of the initiative. The two sources are not compatible, and in the source material available the contradiction is unresolved. It could mean that both essays played a role, or that sources are given different weight in different companies. Until more documentation emerges, readers should treat the origin story as open.

The FINRA comparison is in any case the most informative detail: it suggests a model in which an industry is allowed to regulate itself under government oversight, funded by the industry – an arrangement that requires government approval to have any form of authority.

The context: stalled public-private cooperation and the UN

CXOToday writes, citing the same anonymous sources at The Information, that the idea gained urgency after the three companies' first attempt to form a public-private partnership "fizzled" in encounters with the Trump administration. This is a single-source detail (CXOToday).

Regardless, the news points to a broader pattern: AI leaders have in recent weeks sought out international and national arenas. On 23 September, OpenAI CEO Sam Altman, Anthropic chief Dario Amodei and Hugging Face CEO Clem Delangue addressed the UN Security Council on AI safety (BankInfoSecurity).

In that same context, an industry-led standards body fits: if government lacks the capacity or will to regulate, industry can offer its own scheme. It is precisely this dynamic that critics are reacting to.

The credibility problems and the "pacing" debate

The most visible counterargument is the companies' own safety records. All three labs scored C+ or lower on the Future of Life Institute's AI Safety Index for 2026 (Crypto Briefing). That companies with such low grades on an independent safety assessment should now lead a standards body for their own industry is a legitimate question of credibility and conflict of interest.

The second criticism concerns "pacing" – the idea that the frontier labs are calling for international slowdowns of progress. One of the most-quoted comments comes from Zoho chief Sridhar Vembu, who, according to CXOToday's account, likened Anthropic's calls to slow down to a "brilliant student asking others to stop reading while he or she studies in secret" (CXOToday). This comment, too, is in practice an opinion relayed by CXOToday, not an independent verification of anything concrete, but it illustrates the scepticism among developers who do not share the labs' premises.

Added to this is the question of whether an industry-led organisation can actually oblige anyone to do anything, or whether it will simply become a voluntary branding project. None of the sources describes enforcement mechanisms.

Status and open questions

The status so far is that, according to Crypto Briefing, as of mid-September 2026 there is no formal organisation, no named leadership and no charter (Crypto Briefing). The Information reports that several candidates have been approached about leadership roles, but none has been appointed (BankInfoSecurity).

The open questions that will determine whether the Standards Authority for Frontier AI becomes meaningful or PR:

• Funding and independence. Hassabis's original proposal points toward industry funding, but the details are unclear. Who pays the bill effectively determines whom the body answers to.
• Authority. Will the body be able to certify or halt models, or only issue recommendations? Without formal authority from a state, it is unclear how the standards would be enforced.
• Leadership. No leader is in place. The choice will say a great deal about how independent the body actually becomes.
• Legal status. Lehane says exemptions from competition rules are not needed, but that is so far a company answer, not a regulatory determination.
• The origin story. Whether Hassabis's essay or Amodei's essay lies behind the initiative remains unresolved in the sources.

In short: the coordination among the three labs appears to be real, while the "Standards Authority for Frontier AI" is for now a name for a plan. By the end of 2027, either an organisation will have launched or it will not – and then it will become clear whether this becomes a serious self-regulatory scheme or a single chapter in the industry's PR campaign against public regulation.