Claude AI Finds SAML Security Flaws

Overview

Security researchers at Oblique Security have leveraged Anthropic’s Claude AI to uncover serious flaws in Security Assertion Markup Language (SAML) implementations. These weaknesses could allow attackers to bypass authentication and take over user accounts, highlighting the fragility of XML‑based identity systems.

How Claude Helped Discover Flaws

Instead of reproducing known bugs, researchers built an AI‑assisted testing harness around Claude Opus:

• Threat modeling: Claude explored how different SAML libraries handled signed XML data.
• Parser differentials: The AI identified unusual parser behavior.
• Exploit validation: Attempted to create end‑to‑end exploits based on observed anomalies.

Key Findings

• Signature wrapping attacks: Attackers can move or inject XML elements so a valid signature verifies one object while the application reads identity data from another.
• CVE‑2026‑57580 in Authentik:
  • Exploited XML comments inside a SAML NameID value.
  • Allowed attackers to link their external identity to a victim’s account under certain non‑default matching modes.
  • Fixed in Authentik versions 2026.2.6 and 2026.5.5.
• Other affected projects: PHP litesaml/lightsaml, OneUptime, and Java saml‑client.
• Denial‑of‑service risks: Excessive memory usage triggered by attacker‑controlled XML documents.
• Authentication bypasses: Found in authentication requests, attribute queries, and logout operations.

Why This Matters

SAML is widely used for enterprise single sign‑on (SSO). Secure implementation requires both the identity provider and service provider to interpret XML consistently. When signature verification and application logic diverge, attackers can inject unsigned identity data that applications mistakenly trust.

Defensive Guidance

Organizations should:

• Use mature libraries: Avoid custom SAML implementations.
• Validate signed elements strictly: Ensure applications only trust verified identity data.
• Disable dangerous XML transforms: Reduce parser exploitation risks.
• Impose resource limits: Prevent denial‑of‑service attacks.
• Test for signature wrapping: Regularly audit identity flows for parser differentials.

Expert in the Cloud Insight

Claude’s role in uncovering these flaws demonstrates how AI can accelerate vulnerability discovery. But it also underscores the fragility of legacy XML‑based authentication systems. For identity teams, the takeaway is clear: SAML implementations must be hardened against parser inconsistencies and signature wrapping attacks, or organizations risk account takeovers at scale.