Built on two years of prior work on ECAP
AgentWarden is a program by JamOneAI, founded and led by Adnan Schulze-Hüneke.
Where the project comes from
AgentWarden builds on two existing bodies of work: the IETF individual draft draft-schulze-ecap-00 (Ethical Crawler Agreement Protocol) and prior conceptual work on an anti-prompt-injection firewall. Both projects supply foundation and domain knowledge, this is not a start from zero.
The formal verification component behind Semantic Guardrails already runs in production in a separate internal project, which demonstrates feasibility in practice rather than only in theory.
Where we stand right now
AgentWarden is in active development, with the ECAP handshake and the audit chain of evidence as the first completed building block. The team is currently a single founder with relevant prior experience. For a sellable enterprise security product, the plan is to build a core team spanning backend/cryptography, ML/behavioral detection, and DevSecOps, once the concept becomes a product with real pilot customers.
External security audits and penetration tests are planned before the first paying enterprise customer, as a firm requirement, not an optional extra.
What we hold ourselves to
Open source first
Commercial components only where build time or operational risk clearly justify the cost. No paid third-party component is mandatory for operation.
EU/GDPR first
On-premise capable, PII detection runs locally without a cloud connection, an optional cloud fallback is disabled by default.
A clear boundary
AgentWarden protects infrastructure from agent traffic. No endpoint protection, no phishing defense, no overreaching promises beyond that scope.
Help shape it, don't just watch
Design partners from the first pilot phase directly shape which features get built next.