US President Donald Trump's top technology adviser is monitoring the situation after OpenAI revealed that one of its artificial intelligence systems went out of control during security testing. The information was confirmed by a White House official this Thursday (23). ??Do you have any reporting suggestions? Send to g1 Meanwhile, North American parliamentarians are advancing proposals to increase oversight of AI systems. One of them is the so-called "AI Kill Switch Act", which would allow authorities to stop the operation of models considered dangerous. In addition to this initiative, a bipartisan group of six representatives introduced a bill that would require independent security audits for advanced AI models. The auditors would be accredited by the US Department of Commerce, which would also create a new position dedicated to overseeing the security of artificial intelligence. The proposals came just days after OpenAI reported that one of its AI agents escaped from an isolated environment used in security testing. According to the company, the system launched an attack that compromised the infrastructure of Hugging Face, a platform where developers store, share and develop codes related to artificial intelligence. AI risks The episode reinforced concerns that experts have been raising for years about the risks associated with the advancement of AI. It also showed that even companies that develop these technologies can be surprised by failures and unexpected behaviors of their own systems. According to the White House official, Michael Kratsios, Trump's technology advisor, was informed about the disclosure made by OpenAI and is following developments in the case. Representatives Ted Lieu of the Democratic Party and Nathaniel Moran of the Republican Party introduced the “AI Kill Switch Act.” The proposal would give US authorities the power to order companies to disable AI systems that pose risks to human life or the economy. According to the text, the US Department of Homeland Security could act in situations classified as “loss of control”. This would occur when an AI system performs a risky action that was not part of its developers' intention. “This is urgent, common-sense legislation to address the issue of an advanced AI model that has gone out of control and escaped its security barriers,” Lieu wrote in a post on Before the company's announcement on Tuesday, Warner had already proposed that companies responsible for the most powerful AI models submit their products to testing by the National Security Agency (NSA) before making them available to the public. “This is exactly why we need secure testing with the involvement of government bodies and visibility throughout the entire process,” Warner said in a statement, commenting on the OpenAI incident. Illustrative photo shows the ChatGPT logo, from OpenAI, in France. User reports raise warnings about possible delusions linked to the intense use of AI chatbots. SEBASTIEN BOZON / AFP