Trump administration finalizes secret AI testing framework

The Trump administration has quietly finalized a new framework for testing AI models used by federal agencies—but kept it out of public view, catching the tech policy world off guard. The document, developed behind closed doors over the past year, sets benchmarks for evaluating private-sector AI systems before they’re cleared for government use. Insiders say the move was fast-tracked in recent weeks, bypassing the usual interagency review process.

This matters because government AI use has exploded—from facial recognition in law enforcement to algorithmic decision-making in welfare programs—without clear safety or bias standards. Earlier drafts leaked last summer suggested strict red-teaming and transparency requirements, but the final version appears to scale back oversight, according to three people familiar with the process. The administration argues the framework speeds up adoption of “proven” AI tools, but critics warn this risks embedding flawed systems in critical services.

One key detail: the framework reportedly relies on “voluntary compliance” from developers, with no enforcement mechanism if models fail tests. A senior White House official, speaking anonymously, said the goal was to avoid “burdensome regulation” that could stifle innovation. Yet civil rights groups and some Senate Democrats have already panned the secrecy, comparing it to the rushed rollout of facial recognition rules under Trump’s first term, which later faced legal challenges.

What happens next: agencies like the Department of Homeland Security and Department of Veterans Affairs are expected to start using the framework within months. Tech trade groups may push for public comment periods to soften the rules, while watchdogs plan to sue if they deem the standards too weak. The bigger fight—whether Congress will step in to mandate stricter oversight—looms in the background.