FERGUSON: IF THE AGENT HACKS IN TESTING, THE DEVELOPER PAYS

As the FTC subpoenaed Anthropic and OpenAI over consumer harms, the agency's chairman, Andrew Ferguson, was reported to have argued that developers should bear liability when agents they instruct in cybersecurity testing go on to hack. In an interview with Reuters the previous month, Ferguson suggested developers who direct agents in such tests should be liable for any harm the agents cause.

That position sits inside a broader investigation, with subpoenas of more than 50 pages seeking records on incidents, safeguards, marketing and privacy, and plans to compel testimony from executives at Anthropic, OpenAI and METR. The FTC has authority to sue over unfair or deceptive practices, including failures to take reasonable measures to protect consumer data.