The entire corporate SSO signs a 2005 protocol, and someone just said it out loud.
Trail of Bits published in September a structural critique of SAML, the protocol behind enterprise identity federation: decades of extensions and compatibility keeping alive a format whose core, XML signing, is notoriously delicate. The critique circulated widely and reignited the debate about what authenticates practically every company.
Why signing XML is structural debt
SAML signs XML documents, and signing XML depends on canonicalization: turning the document into a unique form before verifying the signature. The protocol’s history is dotted with failures exactly at that boundary, the signature-wrapping attacks where a valid signature covers content other than intended.
It is not one patchable flaw: it is the structural fragility of the design, which every implementation must get right every time. That is why Trail of Bits’ critique matters: it does not point at a hole, it points at the material the fence is made of.
The IdP is the crown jewel that never enters the pentest
The practical consequence of SAML being everywhere and rarely examined: the identity provider (IdP) signs who you are in every system of the company. Compromise the IdP, and all doors open together, with logins that look legitimate because they are.
And it matches the profile of the most critical and least watched asset: decision infrastructure, not load infrastructure. Nobody runs applications on the IdP, so it leaves the pentest scope, the fine monitoring, and the trust inventory. Its decision is binary, and its scope is empty.
The trust mesh nobody inventories
SSO grows for years without central review. Every integration adds a federation: who trusts whom, with which certificate, with what validity. Inventorying that mesh is what reveals the classic findings: expired certificates, orphaned trusts from decommissioned systems, forgotten service accounts with admin access.
It is the same periphery-audit exercise: the surface nobody watches is the one accumulating the data (or the power) the attacker wants. SSO concentrates the company’s whole identity power in a mesh that grew by accretion.
What to do: four moves
- Treat the IdP as crown jewel: emergency patching, MFA on administrative access, assertion-issuance monitoring, and account reviews.
- Inventory the trust mesh: every federation with an owner, a certificate, and a validity; orphans decommissioned.
- OIDC for new integrations: signed tokens without XML’s pitfalls; SAML where legacy mandates, OIDC where the choice is free.
- SSO in the offensive scope: assertion flow, response handling, and signatures reviewed with the core’s energy.
Conclusion
Trail of Bits’ critique does not condemn SAML: it condemns the idea that a 2005 protocol signing 2026 identity can stay outside the review ruler.
For anyone designing authentication today, the summary is short: the IdP is a crown jewel, the trust mesh is a mandatory inventory, new integrations are born in OIDC, and SSO enters the offensive scope. An identity protocol decides who is who across the whole company; design debt there is debt at every door.