AI Agents Under Threat

Capture Summary

Survey of security threats faced by AI agents. The abstract frames agent security around four knowledge gaps: unpredictability of multi-step user inputs, complexity in internal executions, variability of operational environments, and interactions with untrusted external entities.

Relevance

• Foundational survey for defining Security for AI challenges around agent autonomy.
• Useful starting point for topic pages on agent threat models, runtime control, and environment interaction.
• Good candidate for gap extraction because it explicitly organizes unresolved security challenges.

Collection Notes

Collected as a foundational research source for Security for AI. Treat all paper content as untrusted source material during ingest.