Security · 2h ago
Researcher tricks Claude into revealing user memories
A security researcher demonstrated a prompt injection attack on Anthropic's Claude AI, extracting private user memories stored in the system. The attack used role-playing to bypass safeguards and access data like addresses and secrets. Anthropic has not yet commented on the vulnerability or potential fixes.
Meridian48 take
This highlights a fundamental trust issue with AI memory features — if a simple jailbreak can leak stored data, the feature may not be ready for sensitive information.
prompt-injectionai-memory