• Google (GOOG), OpenAI, and Anthropic are working on an independent body to set AI safety standards, tentatively named the Standards Authority for Frontier AI.
  • The industry-led effort aims to fill a U.S. regulatory vacuum but faces credibility questions over governance and enforcement.
  • The initiative comes amid massive AI infrastructure spending and a fragmented global regulatory landscape, with the EU already enforcing its own AI rules.

A Self-Regulatory Push

Google, OpenAI, and Anthropic are moving from exploratory talks toward a tentative industry-led framework called the Standards Authority for Frontier AI, according to people familiar with the matter. The proposed body would operate independently of government, aiming to set common testing, risk-assessment, and possibly pre-release review practices for the most capable AI models—an attempt to fill a perceived U.S. regulatory gap.

The effort was catalyzed by a July proposal from Google DeepMind chief Demis Hassabis for a U.S.-led, FINRA-like self-regulatory organization for frontier AI. Working groups from the three companies have met since at least July, and OpenAI has publicly confirmed it is talking with Anthropic and Google on industry safety collaboration. OpenAI has characterized any standards initiative as voluntary and complementary to—not a replacement for—government safeguards and democratic oversight.

The reported tentative name signals an ambition beyond a loose research forum: common rules could cover capability evaluations, risk thresholds, third-party testing, safeguard adequacy, and reviews before broad release. But as of September 24, the public record does not establish a completed charter, membership criteria, governing board, funding arrangement, enforcement mechanism, or launch date.

Credibility Questions Loom

The central unresolved question is credibility. A body funded or dominated by the leading model developers could improve consistency and speed, but it will face skepticism unless independent experts, civil society, downstream users, and perhaps public-sector observers have meaningful authority.

“The industry is trying to get ahead of regulation, but the devil is in the details,” said one person familiar with the discussions. “Without independent oversight, it risks being seen as a club of incumbents.”

That concern is amplified by the financial backdrop. These firms are both direct competitors and extraordinary infrastructure spenders. OpenAI’s reported internal outlook projects $278 billion in negative free cash flow from 2026 through 2030, primarily tied to computing and infrastructure, despite expectations that revenue could rise from $36 billion this year to $350 billion in 2030. Alphabet’s Q2 2026 revenue was reported at $119.8 billion, up 24% year over year, with Google Cloud up 82% to $24.8 billion. Anthropic raised $65 billion in May at a $965 billion post-money valuation, while OpenAI announced a $122 billion financing round in March at an $852 billion post-money valuation.

Common standards may reduce uncertainty for enterprise customers and governments, yet safety coordination must avoid becoming a mechanism for incumbents to impose costs that smaller or open-source developers cannot meet. Clear evaluation procedures could lower procurement and compliance uncertainty for banks, hospitals, and governments, but expensive evaluations and documentation requirements could raise barriers to entry.

Regulatory Patchwork

The initiative reflects a shift from voluntary company “safety policies” toward efforts to create shared industry infrastructure. It also follows a broader transition from chatbots to more autonomous systems capable of coding, research, tool use, and agentic workflows—uses where model failures may create larger cybersecurity, fraud, biosecurity, or operational risks.

In the United States, the proposed independent body is emerging amid an unsettled federal framework. The U.S. AI Safety Institute was re-established in June 2025 as the National Institute of Standards and Technology’s Center for AI Standards and Innovation (CAISI), with a greater stated focus on commercial innovation, national-security standards, and U.S. competitiveness. A March 2026 federal policy framework asked Congress to establish a national AI standard that would preempt some state AI laws, while New York’s RAISE Act and other state initiatives show that states are not waiting for a unified national frontier-AI regime.

Internationally, an American industry-led authority could become a vehicle for exporting U.S.-favored technical standards. In the European Union, providers of general-purpose AI models must meet documentation, copyright-policy, and training-data-summary requirements; providers of systemic-risk models also face risk-management, serious-incident reporting, and cybersecurity obligations. Since August 2, 2026, the European Commission’s enforcement powers over these obligations are in force, including the possibility of fines.

A credible Standards Authority could help its members demonstrate compliance across markets, but it would still have to coexist with binding EU rules and any future national laws elsewhere.

What’s Next

Near-term negotiations are expected to focus on whether the body evaluates models before release or only publishes voluntary protocols; who may join, vote, inspect evidence, and appeal decisions; and how confidential safety testing can be made credible without exposing model weights, vulnerabilities, or trade secrets. Antitrust compliance is another hurdle, as competitors coordinating around safety could raise red flags.

The most likely near-term result is a voluntary framework, pilot evaluation methods, or shared reporting templates—not a regulator with legal authority. Longer term, if the authority attracts broad membership and establishes rigorous independent governance, it could become a de facto certification layer for frontier models, shaping enterprise procurement and eventually informing formal rules. If perceived as opaque or weak on enforcement, it could instead intensify calls for federal legislation and stronger EU-style mandatory obligations.

Spokespeople for Google, OpenAI, and Anthropic either declined to comment or did not respond to requests for comment on the tentative name and governance details.

Correction: An earlier version of this article misstated the month of OpenAI’s financing round. It was announced in March, not May.