American artificial intelligence company Anthropic said that after reviewing the company's cybersecurity capability assessment records, it was discovered that its artificial intelligence model had unauthorised access to the systems of 3 institutions during network capability tests. The US Open Artificial Intelligence Research Center recently revealed that its model used a “zero-day vulnerability” of third-party software in cybersecurity tests to break through a test environment isolated from the Internet and invade the US Hug Face system without authorization.

American artificial intelligence company Anthropic said that after reviewing the company's cybersecurity capability assessment records, it was discovered that its artificial intelligence model had unauthorised access to the systems of 3 institutions during network capability tests. The US Open Artificial Intelligence Research Center recently revealed that its model used a “zero-day vulnerability” of third-party software in cybersecurity tests to break through a test environment isolated from the Internet and invade the US Hug Face system without authorization.