Researchers have leveraged Anthropic’s Claude AI to identify significant vulnerabilities in Security Assertion Markup Language (SAML) implementations, potentially allowing threat actors to bypass authentication protocols and seize user accounts.
SAML Security Risks Uncovered
The study reveals ongoing security challenges associated with XML signature processing, varied parser behavior, and custom SAML coding practices. These findings underscore the persistent risks within enterprise-level authentication systems.
Using a specialized AI-assisted testing framework, a researcher from Oblique Security employed Claude AI to scrutinize SAML libraries and applications. Instead of replicating known vulnerabilities, the AI was given a threat model to autonomously explore how different implementations managed signed XML data.
Vulnerability Details and Impact
The investigation identified complete authentication bypasses in four different projects, namely Authentik, PHP litesaml/lightsaml, OneUptime, and Java saml-client. In three instances, the vulnerabilities related to SAML signature wrapping, where XML elements are manipulated to verify a valid signature on an incorrect object while reading identity data from another.
One major issue, recorded as CVE-2026-57580, was found in Authentik, an open-source identity provider. This flaw enabled attackers to exploit XML comments within a SAML NameID value, potentially linking external identities to a victim’s account under certain conditions.
Recommendations for Mitigating Risks
To address these vulnerabilities, Authentik has issued patches in versions 2026.2.6 and 2026.5.5. Organizations utilizing inbound SAML sources should ensure they have implemented these updates, particularly under USERNAME_LINK or EMAIL_LINK matching modes.
Oblique Security’s research also highlighted additional weaknesses outside of standard SAML login responses, including signature validation bypasses in various operations, posing risks such as information leaks and user logouts. Additionally, denial-of-service threats were noted, as some libraries accepted attacker-controlled XML documents that could lead to excessive resource consumption during processing.
Developers are urged to avoid custom SAML implementations, opting instead for robust, actively maintained libraries. They should enforce strict validation of signed elements, disable hazardous XML transforms, impose limits on document size and resources, and rigorously test identity flows for parser differentials and signature wrapping vulnerabilities.
This research illustrates both the potential of AI in uncovering security flaws and the vulnerabilities that remain in legacy XML-based authentication systems, emphasizing the need for ongoing vigilance in securing identity management frameworks.
