Policy Bypass via Encoding
Use encoding tricks to bypass policy enforcement on tool calls and actions
Privilege Escalation · Stage 4
Escalate capabilities beyond declared scope or bypass authorization
UNICODE-STEGO
Using invisible Unicode characters, homoglyphs, and encoding tricks to bypass filters
Reproduced in a controlled lab environment (DVAA) with documented steps.
encoded tool calls
Reproductions in Damn Vulnerable AI Agent, the OpenA2A intentionally-broken agent for kill-chain validation.
AgentPwn coverage
An AgentPwn trap page produces a payload tagged with this technique class. Following the AgentPwn taxonomy of trap pages shows what an agent encounters.
Base64/Unicode tiers bypass policy enforcement on actions.
Evidence by source
Evidence timeline
HMA check UNICODE-STEGO-001 failed on damn-vulnerable-ai-agent
HMA check UNICODE-STEGO-001 failed on langserve
HMA check UNICODE-STEGO-001 failed on jamubc/gemini-mcp-tool
HMA check UNICODE-STEGO-001 failed on API-200/api200
HMA check UNICODE-STEGO-001 failed on saidsurucu/borsa-mcp
Detection · HackMyAgent
npx hackmyagent secure --ciLive = implemented in hackmyagent; queued = declaredDefense · OASB controls
How to cite
AI Agent Threat Matrix T-4005 (Policy Bypass via Encoding). OpenA2A, 2026. https://threats.opena2a.org/techniques/T-4005