AISchneier on Securityabout 2 hours ago
LLMs and Contextual Integrity
3 min read
CIMemories, a new benchmark, found frontier LLMs leak sensitive memory data in up to 69% of attribute-level violations. GPT-5's violations rose from 0.1% to 9.6% as tasks increased. A separate RL framework reduced inappropriate disclosure while maintaining task performance.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
