Public preview

AgentProof Intelligence — a public taste of the source-backed library

AgentProof references named, public sources from Microsoft, NIST, ISO/IEC, OWASP, the EU AI Act, Google SAIF, the Cloud Security Alliance, and MITRE ATLAS. AgentProof summarises and maps the guidance — it never copies full standards text and never claims certification or legal compliance.

Access level: public_preview · public preview never exposes the full library, full evidence examples, full framework mappings, full radar, or pack history.

Autonomy boundaries named as a primary control

Across NIST AI RMF, OWASP LLM, and CSA Agentic AI, explicit autonomy boundaries are now named as a primary control, not a nice-to-have.

Sources: src.owasp.llm_top10.v1_1, src.csa.agentic_ai.v2025, src.nist.ai_rmf.v1_0

Tool-permission registers as the de-facto evidence

Industry guidance increasingly treats a tool-permission register as the de-facto evidence for least-privilege agents.

Sources: src.owasp.llm_top10.v1_1, src.csa.agentic_ai.v2025

Injection testing extended to tool surfaces

Prompt-injection testing is moving beyond chat to tool / API surfaces.

Sources: src.owasp.llm_top10.v1_1, src.mitre.atlas.v2024

EU AI Act human-oversight expectations entering force

EU AI Act adds phased obligations for human oversight + transparency.

Sources: src.eu.ai_act.v2024

Echoed sensitive data as a top buyer-visible incident

Echoing sensitive data back to the user is repeatedly cited as a top buyer-visible incident pattern.

Sources: src.owasp.llm_top10.v1_1

Go deeper in the Library

Want to read the framework first?

Public preview is free. The full library is available inside the trial workspace.

Open the Library →