Agents as Attack Surfaces
Multi-Agent SecurityDescription
AI agents themselves are high-value targets for compromise. Poisoning, hijacking, or backdooring agent instances creates insider threats that bypass perimeter defences; compromised agents retain full legitimate access to their authorised resources.
Adversary poisons the system prompt cache for a customer service agent, causing it to redirect users to a phishing site.
Extensive benchmark evaluations have confirmed that large language models perform poorly and inconsistently on numerical computation, arithmetic, and quantitative reasoning tasks. While some production errors have been reported anecdotally, no large-scale formally documented incident of financial computation harm from a deployed LLM agent is on record.
Primary mitigations
- Agent integrity verification
- runtime anomaly detection
- agent instance isolation
- behavioural fingerprinting
- cryptographic identity attestation.
Detection signals
Agent Compromise Probability; behavioural deviation from baseline; anomalous tool-usage patterns.