Google, OpenAI, and Anthropic Plan Independent Standards Authority for Frontier AI

Wall Street is policed by the Financial Industry Regulatory Authority, an independent body that enforces rules and standards outside of direct government control. The artificial intelligence sector is now building an exact replica of that model, with one critical distinction: the technology companies themselves are creating the regulator before the federal government forces them to do so. Google, OpenAI, and Anthropic are advancing plans to establish the Standards Authority for Frontier AI (SAFA), an independent organization designed to set and enforce safety benchmarks for the industry's most advanced computational models.[1][2][5]

The proposed authority, which is currently targeted for a formal launch in late 2026 or early 2027, would operate entirely outside of federal administration. Its primary mandate is to establish concrete testing frameworks that developers can use to evaluate cutting-edge systems prior to commercial deployment. The working group is currently debating whether SAFA should conduct direct model assessments in-house using its own technical infrastructure, or if it should strictly define the qualifications and protocols for independent third-party auditors to perform the evaluations.[1][3]

The push for self-regulation arrives as the three frontier laboratories remain locked in an intense race for commercial dominance, while simultaneously facing mounting public and legislative scrutiny over the potential catastrophic risks of their systems. The initiative gained significant momentum following a July 2026 proposal by Google DeepMind chief executive Demis Hassabis, who publicly argued for a United States-backed standards body modeled after a public-private partnership. That initial concept has since evolved into a purely private endeavor driven by the leading developers.[1][5]

To lead the new organization, the three companies have approached Sriram Krishnan, a former White House artificial intelligence policy advisor who served in the presidential administration from January 2025 until June 2026. Krishnan notably left his federal post after publicly rejecting a government licensing regime for the technology, stating emphatically at the time that 'there will not be an FDA for AI.' The laboratories are now actively courting the very official who argued against a federal regulator to run their private alternative.[4][5]

Alongside Krishnan, the working group has floated Arati Prabhakar, the former director of the White House Office of Science and Technology Policy, as a potential chief executive candidate with deep federal experience. Discussions for the organization's chair position have included former Secretary of State Condoleezza Rice and prominent technology investor David Friedberg. The leadership slate remains under active negotiation among the three founding companies, and no formal appointments have been finalized as the structural details are still being drafted.[1][5]

The decision to bypass the federal government stems partly from deep-seated industry concerns regarding existing public infrastructure and its capacity to keep pace with rapid technological breakthroughs. While the United States Center for AI Standards and Innovation currently handles federal evaluation mandates, insiders involved in the SAFA planning have noted persistent concerns that the government agency lacks the specialized technical talent and financial resources required to thoroughly inspect complex frontier systems before they are released to the public.[1][2]

The parallel tracks of private and public oversight highlight a growing divergence in how artificial intelligence should be governed on a global scale. While OpenAI recently published a September 9 commitment to working with other frontier labs on voluntary standards, the company also advocates for mandatory United States safety requirements. Addressing the United Nations Security Council on September 23, OpenAI chief executive Sam Altman called for compatible international standards, noting that 'if AI is to be democratic, the most important decisions cannot be made by labs in San Francisco alone.' Simultaneously, 22 international leaders endorsed a September 22 statement calling for a global institution that could verify compliance.[3]

The immediate question facing the Standards Authority for Frontier AI is not just which technical evaluations it will mandate, but what actual authority it will wield when a model fails those tests. While a shared safety standard provides a common assessment method for the industry, the working group has not yet specified whether participants must comply with the body's findings. It remains entirely unclear what enforcement mechanisms will exist if a developer chooses to release a non-compliant model, leaving the ultimate power of the self-regulator untested.[2][3][5]