Skill-Inject on prompt injection in agent skill files accepted at NeurIPS 2026, 80+ citations
maksym_andr · x · 2026-09-25
The paper Skill-Inject has been accepted at NeurIPS 2026. It studies prompt injection attacks embedded in agent skill files, and the poster notes it has already accumulated 80+ citations, calling skill-file injection "a big deal."
The topic is timely for the agent ecosystem: skills/MCP-style mechanisms let agents load external instruction files, and malicious injections there can hijack agent behavior — an attack surface AI security researchers are watching closely.
Related event: Skill-Inject Paper on Agent Skill Prompt Injection Accepted at NeurIPS 2026(2 posts)→
More from Safety
- Leaked docs: Meta raked in $16 billion a year from scam ads and shelved the fix — ben_j_todd · 2026-09-25
- Berkeley kicks off AI & Society season 3 with panel on HF and OpenAI security incidents — berkeley_ai · 2026-09-25
- Cursor randomly requesting Apple Music access confuses users — gaganghotra_ · 2026-09-25
- UK government AI guidance: civil servants told to use Gemini Flash and skip saying thanks — chalmermagne · 2026-09-25
- GitHub Security Lab ships AI-powered fuzzing pipeline that autonomously hunts C/C++ bugs — mboehme_ · 2026-09-25
- Zelle phone-number lookups resolve to full contact profiles, rate-limited at 120/min — MicahBerkley · 2026-09-25