After declaring a largely hands-off approach to AI regulation, the Trump administration will now test and review large-scale AI models pre-deployment.
Why it matters: The new testing framework could signal a more direct approach to AI regulation at the federal level, particularly for frontier AI models.
Background: The Trump administration vowed to accelerate AI innovation and avoid potentially encumbering security guardrails as part of an effort to ensure American AI dominance. The administration rescinded previous Biden-era executive orders and guidance aimed at ensuring security in AI deployment as part of that effort.
Why the change? The recent release of advanced AI models such as Mythos by Anthropic posed significant risks in terms of large-scale cyberattacks, infrastructure threats and other national security risks.
- The Trump administration may also be taking AI security and economic risks more seriously, ahead of midterm elections in November.
New testing framework: The federal government will now conduct pre-deployment evaluations and safety tests of frontier AI models, per a new agreement reached between the Center for AI Standards and Innovation (CAISI, a new division of the Department of Commerce) and Google, Microsoft, and xAI, and updated deals with Anthropic and OpenAI.
The testing and reviews will “better assess frontier AI capabilities and advance the state of AI security,” according to the Commerce Department’s press release.
- “Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” said newly appointed CAISI director Chris Fall in a statement.
- The new measures match much of the previously rescinded Biden-era directives.
The bigger picture: The pivot away from unfettered innovation towards greater oversight reflects growing concerns about newer AI models’ safety. However, it is unclear whether this scrutiny will extend beyond frontier models to enterprise AI deployments.