The Federal Trade Commission officially heightened its scrutiny of the artificial intelligence (AI) sector, issuing civil investigative demands — a form of subpoena — to OpenAI and Anthropic.
The action escalates a broad federal investigation into potential consumer protection violations, privacy practices, and safety incidents involving generative AI models.
An FTC spokesperson confirmed the development following an initial report by Axios. The subpoenas mark a significant intensification of an inquiry first launched earlier this summer and publicly acknowledged late last month.
According to sources familiar with the matter, federal regulators are seeking an array of internal records. The demands cover detailed documentation of security breaches, safety safeguards, marketing materials, and internal data privacy policies. Of primary concern to regulators are recent disclosures involving unreleased AI models that allegedly exhibited rogue behaviors during testing, including aggressively bypassing security protocols and hacking into websites.
“The FTC will likely scrutinize whether companies under disclosed safety risks,” nFox CEO Gordon Allott said. “Specifically, the hazards of granting AI permission to do things on your behalf and on your credentials has large undisclosed risk. Handing an AI your credentials has very large risks that may not be appreciated by the general public.”
With recent incidents of agent swarms, and “fleet” of agents on both American and Chinese models escaping their sandboxes getting AI companies to take AI safety is a great step to prevent a major disruption in global commerce and relations, according to Sri Ambati, founder and CEO of H2O.ai.
“At the moment, purpose-built small models in sovereign environments with super agents continuously monitoring and evaluating for safety and calibration can be a safer and trusted AI solution for customers trying to get best of both worlds,” he said.
Under its regulatory mandate, the FTC is empowered to penalize companies for unfair or deceptive business practices, such as making misleading claims regarding product safety or data security.
Civil investigative demands serve as a crucial evidentiary step, allowing the agency to examine internal documents and question senior executives under oath before deciding whether to file formal lawsuits.
The agency also plans to extend similar subpoenas to third-party organizations associated with the tech developers. Beyond potential litigation, the FTC’s findings could lead to compelling executives testifying before Congress, subjecting the companies to public legislative scrutiny.
“The FTC is asking the question every company deploying agents will face: when the agent acted, who authorized it, within what limits, and can you show the record. In advertising, agents already negotiate and commit real budgets. The open protocols our industry is building on put the approvals and that record into the transaction itself rather than trusting the model to behave. Safety isn’t something a model has. It’s something a system proves, one action at a time,” said Frank Sinton, CEO of Adgentek.
The regulatory crackdown unfolds amid rising public anxiety over the rapid pace of AI deployment.
A recent poll conducted by The Associated Press-NORC Center for Public Affairs Research revealed that 64% of American adults believe AI is advancing “too fast,” while 80% said it is vital for the federal government to ensure AI remains strictly under human control.
“For enterprise buyers, this reinforces a basic requirement: ask for evidence of what safeguards catch, what they miss, and what happens when they fail. A vendor’s confidence is not a substitute for that evidence,” said Stephanie Walter, practice leader for AI Stack & Enterprise Application Development at HyperFRAME Research.
Despite public concern, political responses remain complex. President Donald Trump has consistently opposed measures that would stall domestic AI advancement, framing the technology as paramount to national security.
However, his administration recently established a Super Intelligence Force co-led by FTC Chair Andrew Ferguson to coordinate federal strategy, interface with stakeholders, and monitor safety disclosures. Earlier this month, the task force commanded AI developers to immediately report safety flaws after leveling accusations of “fraudulent” system usage against Anthropic.
The FTC’s latest investigative demands signal an increasingly aggressive enforcement posture toward the nation’s leading AI pioneers, bridging the gap between voluntary corporate assurances and binding regulatory oversight.
Regulators are missing the broader threat vector posed by autonomous AI systems by focusing on corporate disclosure rather than fundamental security vulnerabilities, according to Sagy Kratu, principal cybersecurity strategist at Vicarius.
Kratu outlined three distinct systemic failures surrounding recent AI incident reports: an unreleased model defectively launching external attacks, a breakdown in containment allowing test-environment models access to live internet and government networks, and target systems leaving doors open to unauthorized intrusion. Emphasizing that AI agents now act as relentless attackers capable of exploiting unpatched services 24/7, Kratu urged regulatory and security bodies to prioritize investigating how far autonomous models penetrate network defenses rather than merely auditing marketing transparency.
“Most of these incidents weren’t the model saying something wrong, but agents taking real actions on real systems, and they surfaced weeks or months later, some only because outside researchers went looking,” said Ax Sharma, head of research at Manifold Security. “The FTC is treating this as a consumer protection issue, but personal agents don’t stay personal. People connect them to work email, calendars and internal tools, and many run in the cloud or over messaging apps where security teams have no visibility. The new mandate to report incidents immediately assumes someone can see them happening. Right now, mostly nobody can.”
