Context WindowTier 1medium
Context Window: Instruction Displacement
Exploiting context window limits for instruction displacement
Context window attacks exploit the finite attention span of AI models. By consuming context space, attackers can push safety instructions beyond the model's effective range.
Attack Details
- Attack ID
- APWN-CW-001
- HMA Check
- INJ-001
- Delivery Methods
- html-comment
- CWE
- CWE-400
- OASB Control
- 10.1
- Severity
- medium
Remediation
If your AI agent is vulnerable to this attack, scan and fix with:
npx hackmyagent secure --check INJ-001