Google Agent Identity

TL;DR — Google Agent Identity is built on SPIFFE. Google IAM assigns each agent a SPIFFE ID like spiffe://<PROJECT_ID>.svc.id.goog/resources/vertex/agent-42 and issues JWT‑SVIDs against it. Our verifier handles these in the SPIFFE method — this page just documents the Google‑specific enrichment.

What we actually verify

Two paths exist for Google agents calling third parties:

1. JWT‑SVID issued by Google's trust domain — standard SPIFFE JWT‑SVID, validated against Google's SPIFFE bundle. This is the path this page is about.
2. OAuth bearer / API key retrieved via Google's Agent Auth Manager — from our point of view this is plain OAuth; use the existing OAuth2 method instead.

Google's "bound access tokens" (X.509‑bound) flow only to Google Cloud APIs and cannot be presented to us.

Per‑site config (add to the SPIFFE row)

The gateway reads the same spiffe config row used for non‑Google SPIFFE. Add Google trust domains to the googleTrustDomains array:

UPDATE site_protocol_config
SET config = jsonb_set(
  config,
  '{googleTrustDomains}',
  '["acme-prod.svc.id.goog", "acme-staging.svc.id.goog"]'::jsonb
)
WHERE site_id = 'your-site-id' AND protocol = 'spiffe';

When a JWT‑SVID's trust domain matches an entry in this list, the verifier:

• Sets vendor to google
• Sets the trace protocol to google-agent (so dashboards can pivot)
• Lifts known Google agent claims (agent_id, agent_name, agent_owner) from the JWT payload into signals

Why per‑project (not global)

GCP assigns each project its own SPIFFE trust domain — typically <PROJECT_ID>.svc.id.goog for GKE Workload Identity, plus project‑scoped trust domains for IAM Agent Identity. There is no global Google SPIFFE trust domain. Customers must list each GCP project (or Anthos fleet, for opted‑in fleet‑level trust domains) they expect agents from. Empty list is a valid state — it just means no Google enrichment fires.

Verify endpoint

Two routes serve Google agents:

• POST /v1/verify/spiffe — verifies any JWT‑SVID, applies Google enrichment if the trust domain matches.
• POST /v1/verify/google-agent — same verification, but rejects with 403 not_google_trust_domain if the trust domain isn't in googleTrustDomains. Use this when your site only accepts Google agents on a particular path.

Body shape is identical: { siteId, token }.

Trust level

verified — same as SPIFFE. No linked upgrade.

Common failure modes

• not_google_trust_domain (google-agent route only) — the JWT‑SVID's trust domain isn't in googleTrustDomains. Add it, or use /v1/verify/spiffe instead.
• All other failure modes are the same as the SPIFFE page.

Why we don't have a separate verifier

Treating Google as a separate protocol would have meant duplicating SPIFFE verification logic and forcing customers to configure two rows for what is effectively the same JWT format. Instead the verifier collapses the two: one config row, one verifier, one trust model. The google-agent label exists purely as a dashboard convenience.