- Industry shift. OpenAI, Anthropic and Google have held working-group meetings on an AI self-regulatory body since at least July — a move that signals the three frontier labs are converging on self-governance before governments impose it.
- Model. The proposed body is modelled on FINRA, the financial industry’s self-regulatory organisation, and would test and audit AI models before deployment rather than responding to failures after the fact.
- Catalyst. Google DeepMind’s Demis Hassabis proposed a US-led version of the body in an essay published in July; Anthropic CEO Dario Amodei followed on September 12 with a public call for industry-wide collaboration on safety standards.
Representatives from OpenAI, Anthropic and Google have been meeting on a working-group basis since at least July 2026 to explore the creation of a joint AI standards body, according to sources cited by CNN and The Information. The discussions, confirmed on September 14, represent the most concrete effort yet by the three leading frontier AI developers to coordinate on safety before regulators impose their own frameworks.
The FINRA Blueprint
The proposed body is modelled on the Financial Industry Regulatory Authority, the self-regulatory organisation that oversees broker-dealers in the United States. Under the FINRA analogy, the AI standards body would set pre-deployment testing requirements for advanced models, conduct independent audits, and maintain a set of safety benchmarks that member companies would be required to meet before releasing new systems to the public. This pre-deployment gate distinguishes the model from existing efforts — such as the voluntary commitments signed at the White House in 2023 — which focus on post-release incident monitoring.
The initiative drew its initial shape from a July essay by Demis Hassabis, who proposed the concept as a way for the United States to lead on AI governance without waiting for Congress to act. OpenAI CEO Sam Altman, according to PYMNTS, told an internal company town hall that he favoured “a testing and auditing organisation for the industry.” Anthropic CEO Dario Amodei published a blog post on September 12 calling for AI companies to collaborate on standards, framing it as necessary before the gap between capability and oversight grows wider.
Timing and Industry Context
The working-group meetings predate a turbulent week in AI safety discourse. A former Anthropic researcher resigned publicly, accusing the major labs of failing to behave responsibly on safety. Amodei then proposed embedding third-party watchdogs inside AI companies as a separate measure. The internal tensions at Anthropic and the working-group discussions are parallel developments, not the same initiative, but they share a common premise: that the industry cannot self-govern informally at the pace the technology is advancing.
California’s mandate for third-party AI audits, the first of its kind among US states, adds political urgency to the companies’ calculation. If the industry can establish a credible self-regulatory body with binding pre-deployment tests, it narrows the legislative case for state-by-state or federal rules that might be less industry-friendly. Whether the working group can resolve the competitive tensions between three companies whose business models depend on racing each other to capability thresholds remains the central question.
What Has Not Been Settled
The group has not yet agreed on governance structure, membership criteria, or how the body’s standards would be enforced. It is also unclear whether the body would cover open-weight models from other developers, such as Meta’s Llama series, or focus exclusively on the closed, proprietary systems built by OpenAI, Anthropic and Google. Those structural questions are consequential: a standards body that covers only its founding members would leave most of the deployed AI market outside its remit.