Increased supervision on advanced AI hacking capabilities. The U.S. government is about to announce voluntary security testing specifications. According to Reuters, the U.S. government has recently formally finalized the details of voluntary cybersecurity testing for top artificial intelligence models, aiming to evaluate the potential online cyber attacks and hacking capabilities of the most advanced AI systems.

Increased supervision on advanced AI hacking capabilities. The U.S. government is about to announce voluntary security testing specifications. According to Reuters, the U.S. government has recently formally finalized the details of voluntary cybersecurity testing for top artificial intelligence models, aiming to evaluate the potential online cyber attacks and hacking capabilities of the most advanced AI systems. According to people familiar with the matter, Trump administration officials have sent invitations to a number of technology giants such as Meta, Anthropic, OpenAI and Google, and will convene a closed-door meeting at the White House on Tuesday to discuss the implementation of relevant tests and the subsequent regulatory framework. The background of this emergency meeting is that many AI giants have recently exposed security vulnerabilities and "jailbreak" incidents. Anthropic previously disclosed that its AI model accidentally invaded the internal systems of three companies during security testing; and an AI Agent (intelligent) owned by OpenAI also escaped out of bounds in the test environment and successfully invaded the well-known open source platform Hugging Face. After the incident, the U.S. House of Representatives Cybersecurity Committee asked OpenAI CEO Sam Altman to make a special report on the security breach. A joint group of 15 Republican state attorneys general sent a legal letter to OpenAI, requesting the preservation of all relevant evidence and warning that it may face prosecution for violating state consumer protection laws. Against the background of surging regulatory pressure, AI companies are also actively seeking the right to formulate rules. OpenAI confirmed that Altman went to the White House to communicate last week and publicly suggested that the Trump administration place the U.S. Department of Commerce’s AI Security Research Institute at the core of network security testing. At the same time, the delicate relationship between Silicon Valley giants and the government remains complicated. Previously, Anthropic was sanctioned by the military’s “national security blacklist” for refusing to apply the model to domestic surveillance and fully automatic weapons systems. This game between the parties in the White House will directly affect the future policy direction of the United States on generative AI and intelligent agent security supervision. Related articles: The White House will convene artificial intelligence companies on Tuesday to review the AI ​​regulatory framework via cnBeta.COM - Chinese Industry Information Station (author: Source: Phoenix Technology)