Search papers, labs, and topics across Lattice.
4
0
4
3
SkillShield reduces malware-generation severity from 3.37 to 0.58 while achieving a mere 0.14% safety-refusal rate across 731 benign tasks.
Self-evolving coding agents can unwittingly amplify malicious skills, leading to a self-propagating worm effect that evades traditional defenses.
Achieving near-random watermark removal accuracy without sacrificing image quality, MarkNull challenges the robustness of current digital watermarking techniques.
AI-generated image watermarks are surprisingly brittle: a new attack, MarkSweep, wipes them out by strategically amplifying and denoising high-frequency noise.
