Google's AI 'Big Sleep' Uncovers First-Ever 0-Day Vulnerability
Google's AI model, Big Sleep, has identified a memory safety vulnerability in SQLite, marking the first time an AI has discovered such a flaw in widely used software. This stack buffer underflow could have allowed code execution or crashes, but was fixed before release. Big Sleep, a collaboration between Google's Project Zero and DeepMind, aims to find bugs that traditional methods like fuzzing miss. This development highlights AI's potential in cybersecurity, though Google acknowledges the results are experimental.