You’ve probably noticed the pattern. A massive enterprise client wants to buy your software. The deal is ready to close. Then, procurement drops the bomb: “Does it support SAML?” You nod, thinking it’s just a standard integration. Two weeks later, your engineering team is trapped in a 2 AM debugging hell, parsing XML signatures that somehow validate nothing, while the client’s IT department blames your app for their broken identity provider.
Enterprise SSO isn’t a security feature; it’s a liability transfer.
We’ve all been told that SAML is the gold standard for enterprise authentication. It’s a lie. SAML is a fractal of bad design. The deeper you look, the more layers of broken complexity you find. XML Signature Wrapping (XSW) attacks alone should have killed it a decade ago. You can have a perfectly valid signature, but because of how SAML parses XML, an attacker can inject a fake identity that the system happily accepts as legitimate. It’s not a bug in your code; it’s a mathematically baked-in flaw in the protocol itself.
So why does it still exist? Because enterprise procurement doesn’t care about security semantics. They care about checkboxes. As one industry insider bluntly put it: “If you don’t have SAML support, I can find a product that does.” The market rewards compatibility, not safety. When your only buying criteria is “works with our existing, broken topology,” software vendors are forced to build fragile, over-generalized identity layers just to close the deal.
When your only buying criteria is “works with our broken system,” you guarantee the next system will be broken too.
And if you think we learned our lesson with newer protocols, think again. OAuth2 and OIDC are showing major cracks. They inherited the exact same disease: trying to satisfy every enterprise integration demand until the security semantics become completely ambiguous. The attack surface just gets larger. Implementing all the main authentication mechanisms is hell.
Now, the fractal is expanding again. Enter the era of AI agents. Go to any major tech company today and ask them how they authenticate agents. Ask them what an “agent identity” even is. Watch them sweat. We are about to bolt autonomous, unpredictable AI systems onto an authentication infrastructure that was already cracking under the weight of human users.
We are about to hand the keys of our fragile identity infrastructure to autonomous agents that don’t even know what a checkbox is.
The cycle will only stop when we stop rewarding compatibility-only thinking. We need opinionated software. We need protocols that prioritize secure semantics over universal interoperability. Next time procurement asks if you support SAML, tell them the truth: you support secure authentication, and SAML isn’t it.
FAQ
Q: But SAML is an industry standard. Aren't you just complaining about bad implementations?
A: No. While implementations can be buggy, the core flaw is structural. XML Signature Wrapping attacks exploit the fundamental ambiguity of how SAML signs assertions versus how parsers evaluate them. You can implement the spec perfectly and still be vulnerable.
Q: If I'm building a SaaS product today, what should I do about enterprise SSO?
A: Isolate the blast radius. Support a strict, minimal subset of SAML or push hard for OIDC, but never let the identity layer dictate your core product architecture. Build opinionated boundaries and be prepared to walk away from deals that demand dangerous backchannel topologies.
Q: Is OAuth/OIDC really just as broken as SAML?
A: It's structurally better (no XML parsing hell), but it suffers from the same philosophical rot: scope creep. By trying to be everything to every enterprise, its security semantics have become muddy. And now, bolting autonomous AI agents onto it without a defined 'agent identity' model is pouring gasoline on the fire.