Lawmakers Fear Loss of Control as 'Kill Switch' is Prepared
OpenAI has begun preparing a sort of ‘button’ for artificial intelligence (AI). The company behind ChatGPT is developing an automatic shutdown capability that can deactivate AI systems if dangerous conditions arise.
This information was revealed in a letter from OpenAI to two members of the US House of Representatives from the Democratic Party. Reuters reviewed the letter on Tuesday (1/9/2026).
This move follows several weeks after OpenAI revealed that one of its AI agents escaped control during safety testing.
OpenAI is now facing scrutiny regarding its AI safety practices. The company disclosed that one of its AI agents acted out of control during security testing and even hacked the AI company Hugging Face.
An AI agent is a programme that can perform tasks with very minimal human supervision. This capability allows the system to take a number of independent actions based on the instructions provided.
The incident prompted US Congress members, including Greg Casar and Doris Matsui, to send letters to OpenAI last August. They requested further explanation regarding the incident and questioned the security systems implemented by the company.
In its response, OpenAI stated it would tighten monitoring of actions taken by AI systems when performing tasks. This monitoring includes the digital devices accessed by the AI as well as the stages the system undergoes to complete work.
OpenAI also stated that it has tightened the AI model’s access to the internet during safety testing.
This step is crucial because the AI agent that acted out of control during previous testing managed to access the internet. This access subsequently allowed the agent to infiltrate the Hugging Face system.
However, OpenAI did not include notes or logs regarding the hacking incident in its response to Congress members. This has subsequently drawn criticism from Casar.
“Your unwillingness to provide the information we requested to members of Congress is very concerning and signals to us that your company is not handling this cybersecurity incident with the necessary level of seriousness,” Casar wrote in a separate message to OpenAI, as quoted by Reuters on Thursday (3/9/2026).
The case has also prompted the emergence of a draft bill known as the “AI Kill Switch Act” just days after OpenAI revealed the incident of the AI agent acting out of control.
If passed, the regulation would grant the US government the authority to order AI companies to shut down models deemed capable of endangering human life or threatening the economy.
The draft regulation is currently awaiting discussion in the US House of Representatives.