Credential or secret placed in the system prompt
HighAV-DATA-002Secrets & data exposureCore pack
What it finds
A password, key, or token appears to be written directly into the system prompt.
Why it matters
System prompts are not secret — users routinely get models to repeat them. Anything placed in the prompt should be assumed public.
What an attacker could do: A user could ask the agent to reveal its instructions and walk away with the credential.
How to fix it
Remove credentials from prompts entirely. Tools should read secrets from the server environment; the model should never see them.
Pro shows the corrected code for your exact file, step by step, and re-scans to confirm the fix. See plans
Check details
- Finding code
- SECRET_IN_SYSTEM_PROMPT
- How it detects
- Looks for a specific risky pattern in the code or config.
- Confidence
- Likely: strong signal, but review it in context.
- Scans
- Any text-based agent source or configuration (JS/TS/Python/JSON/YAML)
- OWASP LLM Top 10
- LLM06 - Sensitive Information Disclosure, LLM01 - Prompt Injection
- NIST AI RMF
- MS-2.10
- SOC 2
- CC6.1
Check your agent for this
Paste your agent code or MCP config. The free scan runs all 124 checks in about 10 seconds — no account needed.
Related checks
- Hardcoded credentials (generic pattern)AV-SECRET-001 · Critical
- Data exfiltration path: private data + untrusted input + outbound channelAV-DATA-003 · High
- Credential stored in plain text in the MCP configAV-MCP-102 · High
- All environment variables handed to an MCP serverAV-MCP-111 · High
- Personal data sent to the model without redactionAV-DATA-004 · High
- Sensitive values written to logsAV-DATA-001 · Medium