analiza aiqbench

This story is about AI agents being used to find and chain security flaws, with the AI capability central to the event. The test was authorized, and Anthropic says its core models, source code, and weights were not accessed, so the reported exposure involved business systems and third-party software. The main takeaway for AI readers is that autonomous agents can compress vulnerability discovery to hours, and that an AI lab's security depends on the vendors and administrative tools around its models, not only on the models themselves. The two sides also dispute the scope of the test, so the account is still contested.