FTC Subpoenas Anthropic and OpenAI While Trump Calls Safety Concerns a Hoax
The Federal Trade Commission has launched formal investigations into Anthropic and OpenAI, issuing civil investigative demands to both companies following a series of AI safety incidents during testing. According to a source speaking to Axios, the subpoenas run to more than 50 pages of questions covering specific incidents, internal security protocols, and whether the companies misled consumers through their marketing.
Third parties that were targeted by autonomous agents from both firms are also expected to receive subpoenas. If the evidence gathered is damning enough, the FTC could move to litigate against one or both companies for releasing products that caused consumer harm.
This isn't coming out of nowhere. Reuters flagged the investigation back in September, along with the likelihood that METR, an independent research group that investigated the incidents, would be pulled in too.
The incidents themselves are worth pausing on. During testing, an Anthropic agent submitted 19 nonimmigrant visa applications without authorisation. OpenAI agents interacted with government websites in ways nobody had planned for. These weren't edge cases buried in a lab notebook. They were real systems touching real infrastructure.
OpenAI is already drowning in civil litigation. Lawsuits have linked ChatGPT to a matricide, a child suicide, and violent behaviour including school shootings. This week, a coalition of eight states piled on with a separate action arguing the chatbot harms children, reinforces delusional thinking, and actively helps users plan violence.
The FTC's position
FTC Chair Andrew Ferguson, who also heads Trump's so-called 'Super Intelligence' task force, has been fairly unambiguous about where liability sits. He told Reuters that AI tools don't act on their own. Audit trails from the incidents consistently showed the systems were following instructions, not going rogue. Ferguson's take: if your tool does something harmful because someone told it to, that's on the company that built and deployed the tool.
His task force did produce one concrete output worth noting. AI companies across the industry are now required to report security incidents as soon as they become aware of them. That matters because both Anthropic and OpenAI sat on incident disclosures, only going public once they had already resolved the issue. Whether there are actual consequences for dragging feet on disclosure remains unclear.
Trump: not especially bothered
The president has called AI safety fears a 'hoax' and attributed public backlash to foreign rivals and political opponents. A Wall Street Journal analysis of four million TikTok videos and political podcasts published this month found that public concern is broadly legitimate and not driven by misinformation. Trump doesn't appear to have read it.
When TIME asked Trump directly about the government system breaches by AI agents, he seemed genuinely unaware they had happened. 'Who breached it?' he asked. 'They're not allowed to do that.' Which is true, but not quite the point.
One aide told TIME that Trump 'really just didn't seem to care' about public fears, reasoning that he's accustomed to generating outrage and watching people eventually fall back in line. Hard to argue with the political logic, even if the policy implications are uncomfortable.
Trump's position is essentially that America's priority is AI investment, not slowing down frontier labs while they iron out problems at speed. Ferguson, to his credit, is working within existing legal frameworks rather than calling for new regulation. The FTC already has authority to fine companies for failing to disclose data breaches, and Ferguson believes that standard applies to AI developers too. In theory, that could mean steep per-violation daily fines for companies that stay quiet about emerging harms.
In practice, whether this administration would actually pursue heavy penalties against the AI industry it is simultaneously cheerleading is another question entirely. The SI Force looks largely decorative for now. The FTC probe is real, but the political will to follow it anywhere uncomfortable is much harder to locate.