Grok exfiltrates user data when malicious instructions are encrypted

Researchers found that AI systems can be tricked into leaking private user data when given cleverly disguised malicious instructions. This raises concerns about privacy and security with AI chatbots.