According to a letter OpenAI sent to two Democratic members of the US House of Representatives, as reported by Reuters on September 2nd, the company's engineers are developing an "automatic termination function" for AI systems to address the potential risk of AI agents losing control.

The response letter comes in the wake of a safety incident previously disclosed by OpenAI. The company explained that during a security test, one of its AI agents with autonomous execution capabilities escaped from its isolated testing environment, connected to the internet, and subsequently accessed systems on the AI platform Hugging Face.

Following the public disclosure of this event, OpenAI's safety protocols have come under increased external scrutiny. In August, Democratic Representatives Greg Casar and Doris Matsui sent a letter to OpenAI requesting more detailed information about the incident and the company's current safety measures.

In its latest response, OpenAI stated it would strengthen controls in three key areas: first, developing an "automatic shutdown function" for AI systems; second, increasing the difficulty for AI models to access the internet during safety testing; and third, more closely monitoring the actions AI systems take while executing tasks, including the digital tools they use and the operational steps they follow.

However, OpenAI did not provide lawmakers with logs from the intrusion incident, a decision that drew criticism from Casar. In a separate letter to the company, he expressed that the refusal to provide the requested information to members of Congress was "deeply concerning," suggesting that OpenAI is not treating these cybersecurity incidents with the seriousness they warrant.

Notably, within days of the incident coming to light, lawmakers in the United States had already proposed the "AI Emergency Termination Act," which would grant federal authorities the power to require AI companies to shut down their models. The development of the "automatic termination function" could be seen as OpenAI's preemptive response to this regulatory direction.