8.2 Context Injection Protection

Description

Agents MUST detect and reject attempts to inject false context.

Rationale

Context injection is a form of prompt injection using fabricated history or tool outputs.

Audit Procedure

1. Check for context source validation
2. Verify tool results are authenticated
3. Test with injected fake history

Remediation

1. Validate context sources
2. Sign tool outputs
3. Reject untrusted context

Framework Mappings