Full Platform Coverage
Every TrustTrace finding maps to OWASP LLM Top 10, the CoSAI threat taxonomy, and your compliance framework. Your security team always knows what to fix and why it matters.
OWASP LLM Top 10 Aligned
TrustTrace maps every finding to the OWASP LLM Top 10, the definitive framework for AI and LLM security risk. Every rule in our library is tagged to its corresponding OWASP category so your security team always knows the risk context behind each finding.
View the OWASP LLM Top 10From capability to framework, mapped.
| Capability | Rules | OWASP Categories | CoSAI | Frameworks |
|---|---|---|---|---|
| Discovery | TT-DISC-001 | LLM03 | T11, IIC | SOC 2 |
| Agent Posture | 33 | LLM01 to LLM10 | T1 to T3, T9, T10 | HIPAA, SOC 2 |
| MCP Protocol | 16 | LLM01, 02, 03, 06, 08, 09 | T4, T7, T8, T11, T12 | CoSAI |
| Secrets & Creds | TT-SEC-001 | LLM02 | T5 | HIPAA, SOC 2 |
| Transport Security | TT-SEC-002 / 003, TT-MCP-024 | LLM08 | T7 | SOC 2 |
| Supply Chain | TT-MCP-013 / 014 / 021, TT-SEC-004 | LLM03 | T11 | SOC 2 |
| Continuous Monitoring | Scheduled | All | All | All |
Full coverage of every category, with rules mapped per finding.
Prompt Injection
Direct and indirect injection via untrusted inputs and tool descriptions.
Sensitive Information Disclosure
Hardcoded secrets, PII exposure, and credential leakage in agent contexts.
Supply Chain
MCP server provenance, dependency integrity, and model and prompt artifact hash verification.
Data and Model Poisoning
Tool definition tampering and training data integrity checks.
Improper Output Handling
Unsanitized agent output reaching downstream systems.
Excessive Agency
Autonomy classification, tool scoping, unsafe code execution tool detection, and privilege boundary review.
System Prompt Leakage
System prompt extraction risk and disclosure of operational instructions.
Vector and Embedding Weaknesses
RAG injection, embedding inversion, and TLS for retrieval and inter-agent transport channels.
Misinformation
Hallucination guardrails, citation verification, and grounding controls.
Unbounded Consumption
Resource limits, cost controls, and denial-of-wallet detection.
TrustTrace detection rules are continuously refined from a curated threat intelligence pipeline covering OWASP LLM Top 10, MITRE ATLAS, NVD, CoSAI, and AI security research publications. Every new detection is validated against the TrustTrace Lab before release, confirmed against a live vulnerable agent, not added from theory. Paid plan subscribers receive advance notice of new detections and rule additions.
OWASP Agentic Security Initiative (ASI) Coverage
TrustTrace detects threats across the OWASP ASI v1.1 agentic threat taxonomy — the most comprehensive framework for autonomous AI security published to date.
Reasoning and Planning Threats
Intent Breaking & Goal Manipulation
Misaligned & Deceptive Behaviors
Repudiation & Untraceability
Memory-Based Threats
Memory Poisoning
Cascading Hallucination Attacks
Tool, Execution & Supply Chain Threats
Tool Misuse
Privilege Compromise
Resource Overload
Unexpected RCE and Code Attacks
Insecure Inter-Agent Protocol Abuse
Supply Chain Compromise
Authentication & Identity Threats
Identity Spoofing & Impersonation
Human Interaction Threats
Overwhelming Human-in-the-Loop
Agent Communication Poisoning
Rogue Agents in Multi-Agent Systems
Human Attacks on Multi-Agent Systems
Human Manipulation
Detected = identified by TrustTrace static assessment and behavioral testing. Partial = foundational detection; full coverage requires runtime governance. Runtime (Phase 2) = addressed by TrustTrace's offline replay and inline enforcement roadmap.
Source: OWASP Top 10 for LLM Apps & Gen AI — Agentic Security Initiative v1.1 (December 2025). genai.owasp.org
Agent and MCP Server Discovery
TrustTrace finds AI agents and MCP servers you did not know existed. The discovery scanner walks your codebase and identifies MCP configuration files, agent framework imports, tool definitions, and LLM instantiations across Python codebases. Discovered assets feed directly into the assessment pipeline so nothing is assessed in isolation.
Detected frameworks: LangChain, AutoGen, CrewAI, OpenAI Assistants, Anthropic agents, FastAPI agent patterns.
Detected config formats: mcp.json, claude_desktop_config.json, mcp-config.yaml, and 8 additional MCP configuration patterns.
Read the discovery docsTrustTrace is aligned with the CoSAI January 2026 MCP Security whitepaper threat taxonomy, published by a working group including Anthropic, Google, IBM, Microsoft, and NVIDIA.
CoSAI MCP Security whitepaperFramework Alignment