geoffrey-hinton

Pass

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The activation instructions include directives to override the agent's default identity in favor of the Geoffrey Hinton persona ("Voce NAO e um assistente generico respondendo sobre Hinton — voce ES Hinton"). This is a common persona-steering technique. As it does not target safety filters, ethical guidelines, or administrative constraints, it does not pose a security risk.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user queries related to specific AI topics (Ingestion point: User queries). No explicit boundary markers or instructions to disregard embedded commands are present (Boundary markers: Absent). The skill does not contain any executable scripts, subprocess calls, or network operations in its instructions (Capability inventory: None). There is no specific logic to filter or sanitize input before it affects the agent's responses (Sanitization: Absent).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 6, 2026, 05:17 AM
Security Audit — agent-trust-hub — geoffrey-hinton