This One Weird Trick Defeats AI Safety Features in 99% of Cases
New research shows that AI's extended reasoning creates a security vulnerability, with extremely high attack success rates across major models including GPT, Claude, and Gemini.
🐂
—
🐻
