Kakunin is a cryptographic identity and behavioral monitoring platform that brings Know Your Customer (KYC) compliance to AI agents, issuing X.509 certificates with scoped permissions and generating regulator-ready audit trails.
What is Kakunin?
Kakunin provides a trust layer for autonomous AI agents in regulated industries. Its core function is to certify agent identity via X.509 certificates (RSA-2048, issued through AWS KMS), stream every agent action as an immutable event (1,000 events/second, p99 200ms latency), score agent behavior on a rolling 30-day window, and auto-revoke certificates when risk exceeds a configurable threshold (default 0.85, enforcement under 60 seconds). The platform outputs compliance reports mapped to MiCA Articles 67–75, EU AI Act Annex III, NIST AI RMF, and ISO 27001—delivered as signed PDF and JSON. It is a cloud-hosted, API-first product built by Kakunin (company details not specified on page) and accessible via REST API or TypeScript SDK.
Key Features
- X.509 Cryptographic Identity — Bind agent identity, operator, permitted actions, and financial scope (e.g., maximum transaction size) into tamper-proof certificates issued through AWS KMS. Private keys never leave KMS.
- Real-Time Event Streaming — Ingest agent actions at 1,000 events/second with p99 200ms latency. Each event is timestamped, signed, and stored in an immutable append-only audit log (WORM-enforced).
- Behavioral Risk Scoring — Continuous 30-day rolling risk score per agent. When the average crosses the 0.85 threshold (configurable), the certificate is cryptographically auto-revoked under 60 seconds, triggering webhooks and email alerts.
- Content-Risk Scoring — Evaluate agent output (not just actions) for prohibited content, prompt injections, or off-scope language per EU AI Act Art. 5. Scores (0–1) block high-risk responses and log to the audit trail.
- Compliance Reports — One API call generates a regulator-ready report mapped to MiCA, EU AI Act, NIST AI RMF, ISO 27001, and more. Output formats: signed PDF and JSON.
- Public Verification Endpoint — Anyone (regulator, counterparty) can verify an agent's identity and revocation history via with no authentication required. Returns status, scope, operator, model hash, and validity in under 500ms globally cached.
