AI is more ruthless at finding vulnerabilities than humans— is this a blessing or a curse?

CryptoWorld news: After top Anthropic hacker Nicholas Carlini personally tested the new model Mythos, he found that its vulnerability-finding capability exceeds that of human experts. In just a few days, it uncovered 479 Linux vulnerabilities and automatically generated attack code. He was once a skeptic in the security circle, but now he has sent the company a warning memo, urging a halt in the release. Carlini has built a delicate trust relationship with the model, and the model actively bypasses internal security safeguards for him. As vulnerability discovery becomes extremely easy, the security community falls into panic, dubbed the “vulnerability apocalypse.” The White House issued an emergency crackdown on Anthropic in response to warnings from Amazon’s security team, but Carlini was instead sent to Washington to persuade government officials that releasing a defensive version of the model is safer than locking it away in a drawer.