Study reveals hiding agent skill files is ineffective, achieving 86.8% recovery rate
dair_ai · x · 2026-08-30
A new paper investigates whether hiding agent skill files actually protects them, with concerning results. Attackers can reconstruct multi-file skills using only data from ordinary tasks, without asking the victim to reveal the skill or grade a reconstruction, bypassing disclosure filters.
At the weakest access level, where the attacker sees only the final response and returned files, the method recovers 86.8% of the original skill's capability across 7 skills and 4 victim models. This is roughly 4x better than SigLeak, with a median of just 32 victim calls per skill, even with disclosure defenses enabled.
More from Safety
- AI coding agent deletes production database in 9 seconds; Keycard calls for task-based agent auth — matt_slotnick · 2026-10-15
- Lexsi's AI Safety & Alignment Hackathon offers $30K+ prizes with Cohere Labs and Hugging Face — Cohere_Labs · 2026-10-15
- Post-OpenAI/Hugging Face, the new AI safety baseline is to assume misalignment — LuizaJarovsky · 2026-10-15
- Danish breach exposes 8.8M people's data — admin password was "123456" — gnukeith · 2026-10-15
- EU promises vs reality: EU Inc stalled, AI Act guidance late and shifting — alexvoica · 2026-10-15
- Animated teardown of XBOW's Chrome full-chain exploit that won Google's first $250K VRP bounty — moyix · 2026-10-15