CVE-2026-49478 in Fulcio reveals a trust boundary failure in OIDC discovery that compounds into certificate chain compromise. The core issue: Fulcio's HTTP client automatically attaches ServiceAccount tokens to outbound requests and follows redirects — behaviors inherited from Kubernetes client defaults that are benign within a cluster but dangerous when reaching external, untrusted OIDC issuers. An attacker controlling a redirect can capture tokens, but the worse outcome is JWKS key substitution: a malicious endpoint returns a rogue signing key that Fulcio caches and uses to verify all future certificates. This isn't just token disclosure — it's a persistent backdoor in your trust chain.
The fix in Fulcio 1.8.6 blocks cross-host redirects and restricts automatic token injection. You must upgrade. Beyond patching, audit your OIDC client configurations for the same pattern — the Go Kubernetes client and similar libraries auto-attach credentials by default. Treat discovery endpoints as untrusted until you've explicitly verified the issuer's identity out-of-band. Pin known-good issuers in your configuration, and implement network policies restricting Fulcio's egress to trusted endpoints only.
The real exposure window isn't just the vulnerability — it's the JWKS cache. A planted malicious key remains trusted until cache expiry or manual purging. After upgrading, flush cached JWKS keys before they refresh automatically. This CVE exposes a class of vulnerability baked into how HTTP clients handle external identity providers; other projects using similar patterns are likely exposed.