Skip to main content

The security infrastructure for AI agents.

AI agents are taking actions in your name. Your SIEM does not see them. Your IAM does not understand them. OpenA2A is the identity, observability, and policy layer the agent economy was built without.

Compare OpenA2A with Okta, Auth0, AWS IAM, and othersSee the 12-week pilot for security teams

Example scan output

7 of 67 lines from npx hackmyagent secure (hackmyagent 0.32.0), run 2026-08-26 13:21 UTC in a fresh clone of opena2a-org/damn-vulnerable-ai-agent at commit 19edfbf, exit code 1. The target is deliberately vulnerable, so a low score is the expected result.

Terminal
$ npx hackmyagent secure
  damn-vulnerable-ai-agent  v0.9.3 · cli · 463 files read · semantic capped at 200
  9 critical issues found
  Security  ━━━━━━━━━━━━━━━━━━━━ 2/100
  │ CRITICAL  Remote Fetch Pattern
  │ vulnerable/deploy.skill.md:21
  │ Fix: Remove the curl|sh or wget|sh pattern from this file
  Path forward: 2 -> 100 by fixing 9 critical + 18 high

Repository, on GitHub, opens in a new tab

370,000+
downloads, pulls, and clones, counted per event
npm, PyPI, Docker, GitHub, Hugging Face
290,000+
AI packages indexed in the OpenA2A Registry
43,000+ scanned to date across six package types
1,750,000+
requests logged by the OpenA2A honeypot network
mostly automated scanners and crawlers, counted per request
Apache 2.0
license on all ten tools
every tool repository, as classified by GitHub

Three entry points. One ecosystem.

Identity, scanning, and credential protection. Each tool is open source and works standalone. Together they form the security infrastructure for AI agents. One CLI wraps all three: npx opena2a-cli init runs a read-only assessment, and the same CLI exposes HackMyAgent as scan, Secretless AI as secrets, and AIM as identity.

AIM

Cryptographic identity, audit logs, trust scoring, and capability policies for every agent. Self hosted or AIM Cloud.

  • Ed25519 cryptographic identity
  • 9 factor trust scoring
  • Append only audit log
  • Capability policy enforcement

HackMyAgent

Static, semantic, and adversarial checks across credential leaks, injection vulnerabilities, and MCP misconfigurations.

  • 310 static checks
  • 164 attack payloads
  • Auto fix with rollback
  • OASB benchmark compliance

Secretless AI

Keep API keys and secrets out of AI tool context. Supports Claude Code, Cursor, Copilot, Windsurf.

  • 57 credential patterns
  • Pre tool use blocking
  • Multi backend storage
  • MCP server encryption

Built on open standards.

The tools implement open, vendor-neutral specifications for agent identity, trust, authorization, threat modeling, and governance. Read the specs, build on them, or bring an existing stack.