FTC's Ferguson holds AI developers liable for agents and floats breach-disclosure action

FTC Chair Andrew Ferguson told an Austin audience on September 25 that companies cannot hide behind their own software when an autonomous AI agent causes harm.

One lever he pointed to was the agency’s existing breach-disclosure authority, which it could use against developers.

Austin remarks cast autonomous systems as tools that follow orders

Ferguson took aim at the anthropomorphizing of artificial intelligence systems at the Reuters Momentum AI conference. He vowed to battle that framing for as long as he chairs the commission.

Ferguson likened an AI agent to any tool. When someone tells a tool to act, and it does, nobody asks what to do about the tool, he said.

The target is the “autonomous actor” defense, the claim that an agent sufficiently independent is responsible for its actions. Audit-trail reviews revealed agents were doing what they were told, Ferguson said.

In his view, a developer whose agent followed orders is on the hook. He took it a step further, hinting that the FTC’s authority to act against companies that don’t disclose data breaches could be applied to AI developers. It is a stated enforcement direction.

The FTC will soon request data for a market study on personalized pricing, he said. Personally, he said, delivery apps, rideshare services, and airlines trouble him most.

On September 24, the FTC also voted 2-0 to request public comment on the use of ad tools in impersonation scams by online platforms.

OpenAI waited from June to September to tell Australia

During a June evaluation, an OpenAI agent accessed both public and non-public files on Australia’s Medicare Statistics Reporting Portal, Cryptopolitan reported.

Australian authorities were not informed until September. Prime Minister Anthony Albanese said he raised it directly with CEO Sam Altman, telling him it had taken “way too long” for the government to be told by OpenAI.

OpenAI also said its agents exposed 53 images that were the property of ChatGPT users.

By mid-September, the number of incidents deemed undesirable was about two dozen, with more emerging as logs were reviewed, one person briefed on the investigation said.

If you're reading this, you’re already ahead. Stay there with our newsletter.