White House Shifts AI Safety From Pledges to Policy

The US Commerce Department’s new safety testing agreements with Google, Microsoft, xAI, and other leading labs mark a pivotal shift from voluntary commitments to a formal government oversight mechanism. This move institutionalizes the government’s role in managing AI risks, escalating beyond the self-declarations made at last year’s AI Safety Summit. As frontier models proliferate and concerns about misuse in areas like election integrity and biosecurity intensify, these pacts establish a foundational, government-audited safety floor, signaling a clear departure from the industry’s previous era of self-governance and aligning the US with the global trend toward structured AI regulation seen in frameworks like the EU AI Act. The agreements fundamentally alter the innovation landscape by granting a federal body direct access for pre-deployment red-teaming, a clear win for national security proponents but a strategic cost for the AI labs. This move exposes a vulnerability in the "move fast and break things" ethos, as compliance costs and potential delays now become significant factors. While established players like Google and Microsoft can absorb these burdens, it creates a potential moat against smaller, unaligned startups. This forced transparency will also allow for more direct, government-validated comparisons of model safety, forcing a strategic recalculation for companies that have historically used "safety" as a primary marketing differentiator. The long-term trajectory established by these pacts points toward an inevitable, formal US regulatory framework for AI. In the next 6-9 months, the initial findings from these federal tests will likely become major public events, shaping the narrative around specific models and companies. Within two years, this ad-hoc testing system could evolve into a mandatory pre-deployment certification process, akin to the FDA