Search the Atlas

Search risks, controls, and glossary terms

HighAgenticCredential Theft from Agent ConfigDemonstrated

Configuration-Embedded Credential Extraction

Security & Identity

Description

API keys, passwords, database credentials, and authentication tokens embedded in agent configurations are extracted by adversaries. Agents often require credentials to access their tools, creating high-value credential concentration.

Example scenario

Attacker gains access to agent's system prompt and extracts AWS access keys hardcoded for S3 access.

Real-world evidenceDemonstrated

Prompt injection and inference-time manipulation have been extensively demonstrated against production LLM applications in red-team exercises and public disclosures, with researchers successfully hijacking agents via malicious content in retrieved documents. Confirmed large-scale production incidents causing measurable financial or safety harm remain limited.

Primary mitigations

  • Secrets management systems (not config files)
  • credential rotation policies
  • agent config encryption
  • runtime secrets injection
  • credential access auditing.

Detection signals

Credential exposure events in agent configurations; secrets management compliance rate.

Mitigating controls

5
Dual coverage

Related risks in Security & Identity