AI RESEARCH
AI Agents May Always Fall for Prompt Injections
arXiv CS.CL
•
ArXi:2605.17634v1 Announce Type: cross Prompt injection is the most critical vulnerability in deployed AI agents. Despite recent progress, we show that the prevailing defense paradigm (data-instruction separation) both fails to detect attacks that operate through contextual manipulation and degrades contextually appropriate behavior. We then recast prompt injection via the lens of Contextual Integrity (CI), a privacy theory that judges information flow compliance with contextual norms.