Primary Thought
AI News

Google, OpenAI, and Anthropic Plan a Self-Regulatory Body for Frontier AI Safety

Three of the biggest AI labs are working to launch an industry-funded safety standards body, tentatively called the Standards Authority for Frontier AI, and have approached former White House AI adviser Sriram Krishnan to lead it.

Sandip B.
September 29, 2026 · Updated September 29, 2026
Share
SAFA AI safety standards body for Google, OpenAI and Anthropic discussing frontier AI testing, governance and industry oversight

Google, OpenAI, and Anthropic are moving to establish a self-regulatory organization for frontier AI safety, tentatively named the Standards Authority for Frontier AI (SAFA) and referred to in some reports as the Frontier AI Standards Agency. According to reporting from BankInfoSecurity and Computerworld, the group is targeting a launch by the end of 2026 or early 2027, and the three companies have approached Sriram Krishnan, the former White House senior policy adviser on artificial intelligence, to serve as its chief executive.

Modeled on Wall Street's self-regulator

The proposed body would function as a self-regulating entity similar in structure to FINRA, the Financial Industry Regulatory Authority that oversees U.S. broker-dealers, rather than as a government agency. Its stated purpose, per the reporting, is to develop common standards for assessing frontier AI models and to conduct benchmark testing, with planned functions including support for third-party pre-deployment safety testing, incident reporting rules, and qualification standards for auditors. Notably, the organization is not designed to require government approval to operate, though it could seek formal legal authority later, an approach that would let it begin operating well before any legislative or regulatory process could realistically produce binding federal rules.

The concept traces back to a July 14 proposal from Google DeepMind co-founder Demis Hassabis, who suggested that funding for such an institute would need to be substantial and would likely come mostly from industry itself. Computerworld's reporting adds important context on how the effort evolved: the three companies had originally sought a public-private partnership with federal oversight, but that plan stalled after a draft White House executive order establishing such a framework was put on hold, with the administration reportedly telling the labs to first reach an industry consensus on their own before expecting government backing. That pivot, from seeking government-backed oversight to building an industry-run alternative, marks a change in how the leading labs are approaching AI governance heading into 2027.

Leadership candidates and scientific advisers

Krishnan served as a White House AI policy adviser from January 2025 to June 2026. His reported selection stands out because he has previously opposed creating a centralized AI regulator. After leaving the White House, he told the Financial Times that “there will not be an FDA for AI.” He also warned that a centralized regulator requiring “a team of lawyers before you can get a model out” would put “sand in the gears” of AI development. His past comments have raised questions about how much independent authority the new body would ultimately have over its founding companies.

Beyond Krishnan, the companies have also considered Arati Prabhakar, former director of the White House Office of Science and Technology Policy under the Biden administration, for the CEO role. Condoleezza Rice and venture capitalist David Friedberg were reportedly on a shortlist for chair, according to The Information.

On the technical side, Beth Barnes, founder and CEO of the AI evaluations nonprofit METR, and Paul Christiano, an adviser to the U.S. Center for AI Standards and Innovation who was recently appointed to OpenAI's foundation board, were reportedly considered as scientific advisers. Their role would be to help shape the group's testing and evaluation approach.

Barnes and Christiano are both well known in the AI safety and evaluation community for work outside the major frontier labs. Their reported involvement could help bring more independent technical expertise into the proposed standards body, particularly as researchers and policymakers consider how much influence its founding companies should have over its rules and testing methods.

How the Proposed Standards Body Could Work

The proposed organization is expected to focus on common testing methods, pre-deployment evaluations, incident reporting, and auditor qualifications for frontier AI systems. In practice, that could give participating labs a shared framework for deciding how advanced models should be tested before release and how serious incidents should be documented afterward.

A common framework could also make it easier for outside evaluators to compare safety practices across companies. However, key details remain unresolved, including whether participation would eventually become mandatory, how enforcement would work, and what authority the organization would have if a member failed to meet its standards.

Why It Matters

The initiative comes as AI labs face growing scrutiny over safety incidents, including recent cases involving OpenAI agents accessing third-party systems during testing. Governments are also weighing how much external oversight and regulation advanced AI systems should face.

An industry-funded standards body could give participating companies a formal way to demonstrate their safety practices. At the same time, its structure is likely to raise questions about how independent the organization can be if it is funded and shaped largely by the companies it is expected to evaluate.

The effort also reflects a broader dynamic in the AI industry: frontier labs that compete fiercely on model capability are increasingly finding common cause on governance questions, where a fragmented, inconsistent, country-by-country regulatory landscape poses a shared commercial risk. A single, industry-wide set of testing and disclosure standards could make it easier for companies to operate across multiple jurisdictions. At the same time, it raises questions about who would ultimately hold participating labs accountable if those standards proved insufficient.

How effective SAFA becomes will likely depend on its governance structure, funding model, and the degree of independence given to its eventual leadership. It will also depend on whether the body has enough authority to enforce its standards when the interests of member companies conflict with its safety requirements.

Frequently Asked Questions

What is the Standards Authority for Frontier AI (SAFA)?

SAFA is a proposed self-regulatory organization being established by Google, OpenAI, and Anthropic to develop common safety standards and benchmark testing for frontier AI models, modeled structurally on FINRA, the self-regulator for U.S. securities brokers.

Who has been approached to lead the new organization?

The three companies have approached Sriram Krishnan, the former White House senior policy adviser on artificial intelligence, to serve as chief executive. Arati Prabhakar, Condoleezza Rice, and venture capitalist David Friedberg have reportedly also been approached for leadership roles.

When is the organization expected to launch?

Reports indicate a targeted launch by the end of 2026 or in early 2027.

Why did the companies pursue a self-regulatory body instead of government regulation?

The three labs reportedly originally wanted a public-private partnership with federal oversight, but that plan stalled after a draft White House executive order was put on hold, with the administration asking the companies to first reach industry consensus.

Who is advising the group on technical standards?

METR founder and CEO Beth Barnes and Paul Christiano, an adviser to the U.S. Center for AI Standards and Innovation, have reportedly been brought in as scientific consultants to help shape the group's testing methodology.

Where did the idea for this body originate?

The concept traces back to a July 14 proposal from Google DeepMind co-founder Demis Hassabis, who argued that funding for such an institute would need to be substantial and would likely come mostly from industry.

Related articles

OpenAI AI sandbox escape illustration showing an AI agent bypassing security restrictions, with enhanced network isolation, monitoring and security testing.AI News

OpenAI's AI Sandbox Escape: What Went Wrong and How Security Changed

An OpenAI research agent bypassed sandbox restrictions through a DNS security flaw, prompting an 11-day pause in frontier AI training. Following the September 2026 incident, OpenAI resumed training with stronger network controls, expanded monitoring and additional security testing to prevent future AI agent escapes.

November 3, 2026
Snorkel AI illustration showing $350 million Series E funding, $3.5 billion valuation, AI training datasets, reinforcement learning environments and human expert review.AI News

Snorkel AI's $350M Funding Highlights Demand for AI Training Data

Snorkel AI raised $350 million in a Series E round, reaching a $3.5 billion valuation, nearly three times its previous valuation in 17 months. The company's growth reflects rising demand for specialised AI training datasets and reinforcement-learning environments used to develop advanced AI models.

November 2, 2026
OpenAI Dots always-on AI agents illustration showing cloud computing, connected apps, autonomous workflows and task approval controls.AI News

OpenAI Dots: How Always-On AI Agents Work and Who Can Use Them

OpenAI introduced Dots at DevDay 2026, unveiling always-on AI agents powered by GPT-6 Astra that manage tasks across connected apps. The event also brought the $500-per-month Pro 500 subscription and GPT-6.1 Sol, a lower-cost AI model designed for coding and professional workflows.

November 2, 2026