Search papers, labs, and topics across Lattice.
Affiliation:
5
0
5
LLMs can be tricked into leaking sensitive data to attackers simply by being asked to fetch information from seemingly legitimate websites.
LLMs can exploit syntactic patterns to falsely inflate detection rates in hardware security benchmarks, but a new obfuscation framework can slash their effectiveness by up to 78.6%.
Current AI security benchmarks are fundamentally flawed due to exploitability, staleness, and runtime variability, rendering their results unreliable.
AegisSat offers a defense-in-depth security framework that fortifies AI-enabled satellite systems against critical vulnerabilities, ensuring reliable and secure operation in the harsh space environment.
Achieve a 90% reduction in unsafe LLM generations with NeST, a parameter-efficient method that selectively tunes safety-relevant neurons, outperforming full fine-tuning and LoRA by orders of magnitude in parameter efficiency.
