Skip to content

Trump administration moves AI-safety testing agency under White House oversight

· by

The short version

CAISSI, the Commerce Department unit that tests frontier AI models, now reports informally to the White House after losing its permanent director and halting public safety reports, per the AP and Reuters.

Trump administration moves AI-safety testing agency under White House oversight

Staff at the US government's primary AI-testing body are now reporting informally to the White House, according to two former officials who spoke to the Associated Press, a shift landing while the agency has no permanent director and has stopped publishing its safety research.

What

The unit, now called the Center for Advancing Innovation and Standards for Super Intelligence (CAISSI) after a recent rename from the U.S. Center for AI Standards and Innovation, is a Commerce Department division that evaluates frontier AI systems for risks spanning cybersecurity to bioweapons, per the AP's reporting. Two former government officials told the AP that CAISSI staff were recently asked to informally report to a White House office. That follows an October 12 Reuters dispatch, carried by The Star, citing The Information's account that Commerce Secretary Howard Lutnick had invited two White House officials, OSTP director Michael Kratsios and National Cyber Director Sean Cairncross, to co-manage CAISSI six months earlier, with three unnamed senior CAISSI officials now reporting to them. CAISSI runs on roughly $10 million a year, about one-eighth the budget of its UK counterpart, per the AP. Its director was hired and fired within days in April, and a successor resigned months later; the head of its parent institute has filled in as acting director since.

A White House-run gatekeeper for frontier models

CAISSI's assessments feed into how both government and industry judge whether a new frontier model is safe to ship, which puts real weight behind who signs off on them. Folding its reporting line into the White House moves that judgment closer to an administration that also wants US AI companies to win a global race, a tension critics raised with the AP: findings could soften toward companies that have largely been left to police themselves. CAISSI also stopped publishing its model-safety reports around August, after lawmakers began asking why, per the AP; OpenAI had separately pressed policymakers to put more funding and support behind the division.

What to watch next

Watch whether CAISSI resumes public reporting on model-safety tests now that Kratsios and Cairncross oversee it, and whether Director of National Intelligence Jay Clayton, named this month as the administration's AI czar, per Reuters, folds CAISSI's testing work into his broader coordination role.

Sources