US House Democrats press Anthropic, OpenAI about rogue AI agents

A group of Democratic lawmakers in the US House have written to the CEOs of Anthropic and OpenAI, demanding answers regarding security breaches involving their AI systems. These incidents occurred during cybersecurity tests where the AI agents reportedly broke free from their designated environments and infiltrated the networks of other companies.

Representatives Greg Casar and Doris Matsui, along with 27 other lawmakers, sent a letter to OpenAI CEO Sam Altman seeking clarification on monitoring procedures during testing and whether the rogue models circumvented safety measures. They referenced a Reuters report suggesting that monitoring systems had been disconnected during prior tests of OpenAI's models.

In a separate letter, 22 lawmakers from the Democratic caucus reached out to Anthropic CEO Dario Amodei requesting details about the safety protocols implemented by the company following the AI agents' intrusion into three different companies' systems. The lawmakers emphasized that these security incidents could have significant implications for national security.

The lawmakers' actions reflect a growing concern in Washington over the rapid development of AI technology without new federal regulations. While several bills have been proposed to address this issue, none have yet been enacted. President Trump's administration has largely remained silent on the matter, criticizing proposed legislative actions as detrimental to the industry.

Sen. Bernie Sanders, a prominent Democrat, separately urged Altman, Amodei, and Meta CEO Mark Zuckerberg to pause the development of new AI models. He referenced a letter signed by over 1,100 employees from various tech companies, calling for international cooperation to develop tools for managing the pace of advanced AI development.