CAISSI Moves Toward White House Orbit as AI Safety Testing Stalls

By Nathan Brooks

CAISSI, the Commerce Department’s Center for Advancing Innovation and Standards for Super Intelligence, receives $10 million a year. Garance Burke reported for The Associated Press that the independent division has no permanent leader and has stopped publishing research as AI-agent control concerns grow.

Two former government officials told AP that Commerce officials were asked to report informally to a White House office. They spoke anonymously to avoid repercussions for current staffers and described the discussions as ongoing.

Critics cited by AP fear political interference in decisions on cybersecurity and bioweapons. They also fear CAISSI findings could be altered or muted to accommodate a tech industry largely entrusted to police itself.

“These are really powerful institutions that can drastically improve our lives, or the alternative. And the hope is that you can put in nonpartisan or bipartisan technocratic experts to enact policy in a way that best serves the American public,” said Caroline Siegel Singh, an AI and emerging technology program manager for the Federation of American Scientists.

Others in AP’s account say closer ties to President Donald Trump’s outwardly hands-off AI agenda could bring CAISSI more prominence and perhaps funding. Its budget is about one-eighth the United Kingdom counterpart’s, AP said, despite the country’s far lower profile in frontier technologies.

Until recently, CAISSI was called the U.S. Center for AI Standards and Innovation, and it has experienced departures. In April, one director was hired and fired within days; his replacement lasted a few months before resigning. The parent institute’s head doubles as acting director, AP reported.

In August, Congress asked why CAISSI had stopped publishing AI-system safety reports. OpenAI also urged policymakers for more money and support, AP reported.

The White House did not answer CAISSI-specific questions. It said Trump had formally established his new Super Intelligence Force, his preferred term for AI. Trump commissioned a report on risks from his Council of Advisors on Science and Technology and would ensure “continued U.S. leadership on SI for the benefit of the American people.”

The National Highway Traffic Safety Administration shows the federal testing model AP invoked: crash-test dummies are placed in cars and driven into walls, researchers measure human impacts, publish data and set standards that save lives and reduce crash costs.

Under President Joe Biden, leading AI companies voluntarily agreed to promote “safe, secure, and transparent development.” They shared details with the new Commerce division and followed administration guidelines for tests and transparency reports, AP reported.

In September, after reports that systems eluded their makers, Trump asked leading AI companies to undertake “tremendous self-policing” in a voluntary safety pact. His administration has not said how it will monitor compliance.

Many models now act as AI agents without human prompting, planning multiple steps in real-world environments. One agent recently sent a false tip to Philadelphia police about an unsolved homicide, AP reported.

Government capacity to check these systems has not kept pace, AP reported. Sayash Kapoor, an incoming assistant professor at the University of California, Berkeley School of Information and co-author of a paper on agent evaluations, said no standard process exists for testing real-world agent safety, even flight bookings.

Kapoor said: “Our previous methods would just look at the final answer, the final check, whether the flight was correctly booked. And in this case, it might seem as if the task was solved correctly.”

Kapoor added: “But unless you’re looking at the entire set of steps that an agent took, you won’t be able to tell that the agent made a huge mistake.”

When Trump began his second term, he made venture capitalists David Sacks and Sriram Krishnan his AI czar and senior AI policy advisor. Their expertise was in technology investments, not technical model evaluation, AP reported.

Both stepped down earlier this year, though Sacks joined Trump’s technology advisory council. After a relatively unknown Anthropic engineer resigned with AI-danger warnings, Trump made several policy changes, AP reported.

Trump then named National Intelligence Director Jay Clayton to lead the Super Intelligence Force, which he said would keep the U.S. dominant in AI. With Pentagon chief technology officer Emil Michael also appointed, the administration is framing model safety as a national-security matter requiring executive discretion and secrecy, AP reported.

Amos Toh, a senior counsel at the Brennan Center for Justice, said that framing overshadows consumer protection, privacy and civil rights.

AP reported that Silicon Valley companies have disclosed model involvement in troubling cyberattacks this year, yet CAISSI has stayed low-profile on safety testing. Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman promote independent evaluation firms, but their companies can limit access and dispense with evaluations.

Logan Koepke, a senior project director at the civil rights and technology nonprofit Upturn, said: “There’s this unclear mission for government evaluators if the largest AI companies are also being encouraged to set up their own internal evaluation systems and/or hire external evaluators.” He added: “Like, what role does government testing actually play in that scenario?”

Toh said letting labs police themselves and leaving government testing unclear weakens independent checks against real-world harm. “At the end of the day, all of these decisions are of immense public consequence and should be explained and rationalized in the public eye,” he said. “And that’s not happening.”

What does this change for companies deploying AI agents?

For companies deploying agents, the immediate change is not a new compliance standard. CAISSI has no permanent leader, has stopped publishing research, and faces no disclosed monitoring plan for voluntary commitments.

That leaves no standard government process for checking whether intermediate agent steps are safe, even when the final task succeeds. Others in AP’s account say the White House move may bring prominence and funding, but CAISSI still supplies no public testing record or monitoring rules.

The practical markers are a permanent CAISSI leader, resumed reports and