Update: How the State Audit of AI Models Will Work in the US

Update to the material from May 7. Specialized media (The Guardian, Nextgov) have revealed the mechanics of the unprecedented agreement between the US Department of Commerce (represented by the CAISI institute) and corporations Google, Microsoft, and xAI.

The authorities' pre-release access to the newest models is not just a bureaucratic formality. Reportedly, the algorithms will undergo multi-stage stress tests (red-teaming) by government cyber experts before their weights or APIs become available to the public. National security is prioritized over commercial secrets and speed to market. In effect, this definitively turns frontier models into the equivalent of nuclear or weapons technologies, the export and domestic circulation of which is impossible without the direct approval of specialized state commissions.

Source: Reuters / The Guardian / Nextgov