“Low-Key” FTC Investigation Into Developers Gets More Visible; Is the US Government (Slowly) Changing Course on AI Safety?

An FTC investigation that few were even aware was going on is reportedly escalating to issuing subpoenas to leadership and key figures at the frontier developers. While developers are not yet being told to stop or even slow down in any formal way, signals coming from the Trump administration appear to be at least a little more mixed on AI safety.

A little more clarity on opaque AI safety probe (though still not enough)

It has been a busy and somewhat odd week to close out September and open October in the AI safety world. It started with a White House meeting involving nearly all the major figures in US frontier AI development, in which the Trump administration appeared to double down on voluntary industry self-regulation (and coined the term “super intelligence”). But now word has emerged that there has, very quietly, been an FTC investigation into these developers going on since even before the Hugging Face attack and that investigation will now move to making formal demands of these same leading AI figures.

These demands will include the turning over of internal information to the government, as well as mandatory appearances for testimony by leadership and key figures. The FTC investigation will employ “civil investigative demands” (CIDs), a form of subpoena often used in conspiracy and RICO investigations that allows for these things to be obtained without any formal charges first being made. These will target the obvious frontier developers that have been having rampant “misalignment” escapades throughout 2026, but also third-party “effective altruism” auditor METR.

An inside source speaking to the media said that the AI safety campaign is not yet ready to tell developers to “stop” or “do anything” meaningful, however; it appears to just be entering a more aggressive information-gathering phase, even as new models are released and new incidents continue to crop up.

FTC investigation to emerge more publicly in coming weeks

The AI safety situation in the US remains confusing. To date, the developer industry has seemed more interested in regulation than the administration that is supposed to be doing it. That could be chalked up to setting terms that are favorable to the large players at the expense of the small, a criticism that METR has been centrally targeted in. But the Trump administration has been consistently “pedal to the metal” on the issue, brushing away public polling that shows even the Republican voter base is broadly against lax regulation and the attendant proliferation of data centers in communities.

The FTC investigation is currently hiring new staff to conduct the CID probe, something that does indicate a new level of seriousness. Trump also recently told media outlets that AI developers can also begin at least expecting the possibility of referral to the DoJ for charges if their “misaligned” models end up attacking critical infrastructure. To this point, the FTC official that spoke to the media anonymously echoed what a lot of critics from the security industry have been saying lately: there are laws already on the books that could be applied to developers that let their models get out of control, such as the Computer Fraud and Abuse Act, and this may now be a reality they have to account for.