Called SAFA, this standards body for advanced models would be independent of the State and led by a public policy figure.
The three major US laboratories are working to create a common structure to define safety criteria for advanced models. The project, which is still in draft form, is tentatively referred to as the Standards Authority for Frontier AI (SAFA) and is intended to launch in late 2026 or early 2027.
The model claimed is that of FINRA, the self-regulatory body of the US financial sector: a private entity that enacts technical rules and controls its own members, outside the direct supervision of the state. In concrete terms, SAFA would establish evaluation benchmarks shared between competing laboratories, and could conduct technical tests before a model is placed on the commercial market. Several names are circulating for management, including Sriram Krishnan, former White House AI advisor, and Arati Prabhakar, former head of the scientific office of the Presidency.
The objections are predictable and relate less to the idea than to architecture: who finances, who governs, what place for outside researchers, and especially what happens when an unfavourable assessment falls on the product of a founding member. The initiative comes at a time when twenty-six prosecutors general are calling for a public regulator in Congress — two competing responses to the same regulatory vacuum.
Read the full article on the original source
Go to the source →