Protum Agent Directory: AI Agent Trust Scores and Security Catalogue
A searchable catalogue of more than 150 AI agents scored and assessed by the Protum team. Every agent is rated 0 to 100 across four security pillars, mapped to OWASP Agentic AI Top 10, the Protum Agentic Risk Matrix (42 risks), and IMDA AI Verify.
Four security pillars
Permissions (what the agent can access and do), Oversight (human-in-the-loop controls and escalation paths), Data Protection (how the agent handles sensitive data and PII), and Supply Chain (the trustworthiness of the models, tools, and APIs the agent depends on).
What the trust score means
A score of 0 to 100 summarises the agent's posture across all four pillars. Scores above 70 indicate a well-governed agent with documented controls. Scores below 40 indicate significant gaps that require remediation before enterprise deployment. The methodology is published at /trust-methodology.
Coverage and inclusion criteria
The directory covers commercial AI agents, open-source agent frameworks, and domain-specific AI products. Inclusion requires that an agent is publicly documented, that its architecture is sufficiently described for scoring, and that it has a meaningful enterprise use case. Read the full inclusion criteria at /agent-directory/overview.
Vendor claims and disputes
Vendors can claim their listing to provide additional documentation and context. Disputes about scores or factual errors are handled through the Right of Reply process at /right-of-reply. Protum publishes the outcome of all disputes.
Browse the Agent Directory