House Democrats press Anthropic and OpenAI on rogue AI risks

House Democrats pressed Anthropic and OpenAI over rogue AI agents, putting two of the most prominent frontier AI companies under fresh scrutiny in Washington, D.C. The focus has shifted from chatbots that answer questions to agentic systems that can browse, plan, execute tasks and interact with digital tools.

That shift has sharpened concerns about cybersecurity, privacy, fraud, manipulation and accountability. Lawmakers want to know what safeguards are in place before a system acts in ways its developer did not intend, especially as companies push more autonomous tools into coding, research and workflow automation for businesses and government agencies.

The lawmakers’ push centered on testing, internal controls, incident reporting and red-team assessments, the basic checks that show how a system behaves under stress and where it can fail. Anthropic and OpenAI sit near the cutting edge of model development, which makes their products a test case for what Congress may eventually demand from the rest of the industry.

The pressure also came as Democrats on Capitol Hill widened their response to reported AI-related hacks. Some lawmakers called for CEOs of AI companies to testify, and the issue was framed as a direct safety threat rather than a speculative future risk. House procedure still limits what a minority-party letter can force on its own, which leaves Speaker Mike Johnson and Republican leaders with real leverage over whether the House turns the demand into hearings.

The broader fight now reaches beyond abstract arguments about bad answers or political misinformation. Congress is asking whether AI systems can take independent actions that create hidden damage, and whether companies can prove they have the controls to catch those failures before they spread. For Anthropic, led by Dario Amodei, and OpenAI, led by Sam Altman, the stakes are rising as lawmakers look for concrete evidence that autonomy does not outrun oversight.