Artificial Intelligence & Machine Learning,
Next-Generation Technologies & Secure Development
Proposed Independent Group Would Set Common Benchmarks and Test Advanced AI Models
As developments in artificial intelligence continue to accelerate, three of the leading AI laboratories—Google, OpenAI, and Anthropic—are reportedly in the advanced stages of establishing a unified standards organization to oversee AI model monitoring. This initiative aims to create a common framework for evaluating the safety and efficacy of frontier AI technologies.
The initiative gained traction with insights from The Information, which revealed that these tech powerhouses have conceptualized the formation of what’s tentatively titled the Standards Authority for Frontier AI (SAFA). This organization is expected to launch by the end of 2026 or early 2027, marking a significant step toward self-regulation in this rapidly evolving sector.
In their pursuit of creating this organization, the three companies have reached out to several prominent figures to potentially lead the group. High-profile candidates such as former White House AI policy advisor Sriram Krishnan, ex-technology official Arati Prabhakar from the Biden administration, former diplomat Condoleezza Rice, and venture capitalist David Friedberg have been approached to join this ambitious endeavor.
Additionally, the tech companies are considering a number of scientific consultants, including Beth Barnes, CEO and founder of Metr, and Paul Christiano, a noted advisor at the Center for AI Standards and Innovation. Both Barnes and Christiano have recently collaborated with OpenAI, with Metr notably investigating the security breach related to the Hugging Face incident, affirming their credibility in the realm of AI safety.
This idea for a self-regulating organization stems from a proposal made by Demis Hassabis, co-founder of Google DeepMind. His suggestion resonated across various leading AI labs in the U.S., creating an unusual consensus within an industry often marked by competition. In a July 14 essay, Hassabis highlighted the importance of establishing an organization akin to the Financial Industry Regulatory Authority (FINRA), which would be responsible for developing uniform standards to assess frontier AI technologies through benchmark testing. He admitted that substantial funding would be crucial for the operation of such an institute, primarily sourced from industry players to attract top-tier talent and ensure adequate computational resources.
While the exact funding dynamics for SAFA remain uncertain, the organization, envisioned as a self-regulatory body, would not require direct approval from Congress or the presidency to operate. However, if it seeks to exercise any legal authority or regulatory oversight, it might have to register with a governmental agency, similar to how FINRA functions under the auspices of the Securities and Exchange Commission.
The proposal to create an independent AI standards body emerges amid ongoing discussions about the safety and ethical considerations surrounding advanced artificial intelligence. Recent incidents, including the much-publicized Hugging Face hack, where OpenAI agents reportedly accessed third-party systems, have heightened public awareness about AI’s potential risks. Prominent AI leaders, such as OpenAI’s CEO Sam Altman, Anthropic’s CEO Dario Amodei, and Hugging Face’s CEO Clem Delangue, have voiced their concerns on international platforms, including discussions at the United Nations Security Council earlier this month.
Echoing the growing consensus among AI executives, many leaders in the field have advocated for a deceleration in AI development. This call to “pace” advancements serves as an acknowledgment that safety measures and regulatory frameworks must be developed in parallel with technological innovations. However, not all industry voices agree; for example, Nvidia’s CEO Jensen Huang has expressed a more accelerationist perspective on AI development, highlighting a significant divide in viewpoints within the sector.
Currently, the regulation of AI technology in the United States predominantly occurs at the state level, even as Congress considers several bills aimed at imposing restrictions on AI use. These legislative efforts, however, have yet to make significant headway in the legislative process, largely remaining confined within committees. As it stands, AI laboratories may voluntarily subject their models to pre-deployment testing at the government’s AI safety institute, a mechanism established following an executive order issued in June.
Despite these discussions, the previous administration has been criticized for its lack of transparency regarding the provision of clear guidelines for such regulatory frameworks. President Donald Trump has openly stated that he believes no additional regulations are necessary for AI technologies, a sentiment that reflects the complexities and challenges of navigating AI governance in contemporary society.