OWASP 2026 Defense Matrix

Filter AgentShield modules by OWASP 2026 category or defense stage.

OWASP ID	Module	Enforcement
`LLM01:2026`	PromptInjectionDetector	Prompt Injection — Direct override, DAN jailbreaks, ChatML delimiters, Base64/Hex decoding, zero-width Unicode steganography.
`LLM08:2026`	PromptInjectionDetector	Untrusted Context Poisoning — Blocks covert HTML comments / RAG directives and restricts non-user context escalation.
`ASI02:2026`	AgentShieldPolicy	Agentic Tool Misuse — Deep argument inspectors for shell chaining, path traversal, metadata SSRF (`169.254.169.254`), and stacked SQLi.
`LLM06:2026`	AgentShieldPolicy	Excessive Agency — Declarative `allow` / `deny` / `require_approval` rules with human-approval callbacks.
`LLM02:2026`	DataLeakageDetector	Sensitive Info Disclosure — Redacts API keys, PEM keys, JWTs, DB URIs, SSNs, and Luhn-valid credit card numbers inline.
`LLM05:2026`	DataLeakageDetector	Improper Output Handling — Strips executable `<script>`, `javascript:`, and inline event handlers.
`LLM07:2026`	DataLeakageDetector	System Prompt Leakage — Detects verbatim system prompt probes and system canary token exfiltration.
`LLM10:2026`	Prompt + Policy	Unbounded Consumption — Enforces input context size budgets, argument length limits, and recursive loop blocks.