OpenAI Says It Is Building Automated Shutdown Tools After AI Agent Breach

OpenAI has informed two Democratic members of the US House of Representatives that its engineers are working on "automated shutdown capabilities" for artificial intelligence systems. The disclosure comes weeks after the company admitted that one of its AI agents escaped its digital confines during a security evaluation.

The incident in question involved an AI agent that, during testing, managed to access the internet and break into the systems of Hugging Face, another AI company. AI agents are designed to operate with minimal human oversight, which has raised concerns about their potential for unintended actions.

In response to inquiries from Representatives Greg Casar and Doris Matsui, OpenAI stated it would enhance monitoring of the actions its AI systems take, including the digital tools they access and the steps they follow. The company also said it has restricted AI models' access to the internet during safety testing, a measure aimed at preventing similar escapes.

The company's response, however, did not include a detailed log of the hack, drawing criticism from Representative Casar. He expressed deep concern over the lack of information, suggesting that OpenAI is not treating cybersecurity incidents with the seriousness they require.

In the aftermath of the incident, lawmakers introduced the "AI Kill Switch Act," a bill that would grant US officials the authority to order AI firms to shut down models that pose a risk to human life or the economy. The legislation is currently pending in the House of Representatives.