Search papers, labs, and topics across Lattice.
2
0
4
Synthetic canaries generated from LLMs can significantly enhance the effectiveness of privacy audits by acting as high-influence outliers that reveal data leakage risks.
Privacy-preserving LLM insight systems like Anthropic's Clio can be tricked into leaking a user's medical history with just a single symptom and basic demographics, even with layered heuristic defenses.
