research

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). Phase 2/4’s runtime WebSearch/WebFetch/WebFetch-like retrieval explicitly fetches public web pages and other third-party sources (outsider-authored free text) and then “Extract”s that readable content into the agent’s LLM context.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)