SAML: A fractal of bad design(blog.trailofbits.com)
342 points by aray07 21 days ago | 178 comments
tl;dr: SAML, the XML-based SSO protocol that powered the rise of Okta, Ping, and OneLogin, is buckling under its own complexity—particularly around XML canonicalization, enveloped signatures, and a bloated "kitchen-sink" spec that has enabled a steady stream of authentication bypass vulnerabilities through 2025. The author argues SAML's design assumptions (transport independence, disconnected IdP/SP topologies) are obsolete in an HTTPS-everywhere, zero-trust world, and that OIDC—which grew organically via focused RFCs—covers every real use case. Recommendation: vendors should stop onboarding new SAML integrations and migrate customers to OIDC.
HN Discussion:
  • ↑Shares horror stories reinforcing that SAML implementations are dangerously broken
  • ~Blames XML-era thinking for SAML's flaws but warns OIDC has its own emerging cracks
  • ↓SAML still has enterprise features (like IdP-initiated flow) that OIDC lacks, so both are needed
  • ↓Article is one-sided; OIDC/JWT have comparable vulnerability classes that weren't examined
  • •Offers technical fixes or pragmatic paths forward (subset support, avoiding canonicalization, libraries)