A risk score is a guess. This API returns a signature.

Ask a scoring-based human verification or bot detection API whether a real person is behind a request and you get a number — a confidence score, assembled from device fingerprints, mouse movement, timing and network reputation. It is a useful number. It is also a negative inference: an estimate of how much the traffic resembles automation.

That framing has a structural problem. Every signal a scoring system relies on becomes an optimisation target for whatever is trying to get past it, and the models doing the trying improve faster than the heuristics catching them. The score has to keep moving. You inherit that movement.

5arz answers a narrower question and answers it positively: was a real human present for this specific session? The response is an ES256 credential — a signed statement, not a probability. It validates against a public key or it does not, and anyone can run that check without asking us.

Two different questions

These are complementary, not competing. Edge bot mitigation should keep doing the job it is good at. It just cannot produce the artifact below.

The last row matters more than it looks. A scoring system must return an answer for every request, including the ones it has no basis to judge. An attestation layer is allowed to say nothing — and "no credential was issued" is a cleaner input to a decision than a confident-looking 0.5.

One call to issue. One local check to verify.

Register, verify, check. The third step needs no network call to us after the first key fetch.

# 1 · Register — no card, no sales call. An agent can do this itself.
curl -X POST https://api.5arz.com/api/agents/register \
  -H 'Content-Type: application/json' \
  -d '{"name":"Acme","contactEmail":"dev@acme.com","mode":"test"}'
# → { "apiKey": "arz_test_…" }   test-key credentials are marked test:true

# 2 · Verify a human — returns a signed PoHF credential (an ES256 JWT)
curl -X POST https://api.5arz.com/api/agents/verify \
  -H "Authorization: Bearer $ARZ_KEY" \
  -H 'Content-Type: application/json' \
  -H 'Idempotency-Key: 7f3a…' \
  -d '{"sessionId":"sess_…"}'
# → { "ok": true, "attestationId": "…", "pohf_jwt": "eyJhbGciOiJFUzI1NiIs…",
#     "jwks_url": "https://api.5arz.com/.well-known/jwks.json",
#     "verify_hint": "Import the ES256 JWK from jwks_url, validate pohf_jwt…",
#     "creditsRemaining": … }

# 3 · Anyone verifies it — offline, any language, any JWT library
#     Public keys: https://api.5arz.com/.well-known/jwks.json

Fetch the JWKS once and cache it. After that, validation is a local signature check — no round trip, no rate limit, no dependency on our uptime. Re-fetch on key rotation.

Your agent can integrate without you

5arz is listed in the official Model Context Protocol registry. An AI agent can discover the server, register itself, verify a human and check the resulting credential without a person opening a dashboard.

Endpoint: https://mcp.5arz.com (Streamable HTTP), plus a plan-management tool. Live tool list and copy-paste client config at 5arz.com/mcp.

Where teams put it

What we do not collect

The credential is deliberately thin. It attests that a person was present for a specific action — not who they are.

5arz stores no biometric templates. Liveness runs on the person's own device and only a signed attestation leaves it. Where a flow requires document-based identity verification, that step is carried out by a regulated third-party identity provider under that provider's own retention policy. A relying party receives proof that a real human was present — and does not receive, and cannot derive, that person's identity documents, history or score.

This is a design constraint, not a feature toggle. A verification layer that quietly accumulates identity dossiers becomes a liability to everyone who integrates it, including us.

Questions

Stop scoring. Start proving.

Two fields, no card, no sales call. Or let your agent register itself through MCP.