{"id":146102,"date":"2026-08-20T12:26:08","date_gmt":"2026-08-20T12:26:08","guid":{"rendered":"https:\/\/www.europesays.com\/ai\/146102\/"},"modified":"2026-08-20T12:26:08","modified_gmt":"2026-08-20T12:26:08","slug":"claude-ai-finds-authentication-bypass-flaws-in-multiple-saml-implementations","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ai\/146102\/","title":{"rendered":"Claude AI Finds Authentication Bypass Flaws in Multiple SAML Implementations"},"content":{"rendered":"

Multiple critical vulnerabilities in SAML implementations after employing Anthropic\u2019s Claude Code in an AI-assisted vulnerability research <\/a>pipeline. <\/p>\n

Security researcher Eric Chiang, the CTO of Oblique Security, investigation uncovered full authentication bypasses, signature-validation flaws, information disclosure risks, arbitrary logout issues, and denial-of-service conditions across several open-source SAML products.<\/p>\n

Claude AI Finds Authentication Bypass Flaws<\/p>\n

Chiang\u2019s research focused<\/a> on the long-standing complexities of SAML, particularly those related to XML Digital Signatures. Many SAML implementations fail when different components parse or canonicalize XML in different ways, allowing attackers to manipulate data that an application erroneously accepts as cryptographically verified.<\/p>\n

Chiang developed a multi-agent environment using Claude Opus<\/a> after joining Anthropic\u2019s Cyber Verification Program. Instead of directing the model to replicate known vulnerabilities, he provided a threat model. He allowed the system to explore target codebases for unusual XML and signature-processing behaviors.<\/p>\n

The workflow consisted of two stages:<\/p>\n

A \u201cgadget\u201d phase to identify potentially dangerous parser, canonicalization, or XML-processing behavior.<\/p>\n

A \u201cfindings\u201d phase to combine confirmed gadgets into end-to-end proof-of-concept exploits.<\/p>\n

Intermediate findings were stored as JSONL records, enabling the pipeline to prioritize leads, eliminate duplicate work, and reject irrelevant issues before costly exploit validation.<\/p>\n

One notable discovery involved the Node.js xml-crypto ecosystem\u2019s handling of XML processing instructions. The resulting differences in canonicalization led to an email truncation scenario in OneUptime, where specially crafted XML could alter the interpretation of a signed SAML NameID by downstream application logic.<\/p>\n

Authentication Bypasses Found<\/p>\n

Chiang reported full authentication bypasses in four projects:<\/p>\n

ProjectIssueAuthentikA comment injection in a SAML NameID could truncate an identity value and enable account impersonation; tracked as CVE-2026-57580litesaml\/lightsamlSignature wrapping affecting SAML Response handling; tracked as CVE-2026-63182OneUptimeSAML Response signature-wrapping weaknessesJava saml-clientSAML Response signature-wrapping weakness<\/p>\n

Signature wrapping attacks generally exploit ambiguity between the XML element verified by a signature library and the element processed by application code. <\/p>\n

If an attacker can maintain a valid signature over one assertion while causing the application to handle a separate malicious assertion, it can lead to a \u201clogin as anyone\u201d scenario. <\/p>\n

The bug in Authentik received additional attention because eight independent researchers disclosed the same issue simultaneously, highlighting how AI tools may be accelerating vulnerability discovery within the security community.<\/p>\n

In addition to authentication flaws, the research identified significant exposure to denial-of-service attacks. The Go xmldsig library received a fix for quadratic memory allocation issues during signature verification. <\/p>\n

At the same time, JavaScript\u2019s xmldom reportedly faces similar memory allocation concerns that affect Node-based SAML libraries. <\/p>\n

Python SAML packages were also found to allow risky XML signature transformations <\/a>via libxmlsec1, including XSLT-based transformations that can produce extremely large documents from unauthenticated requests.<\/p>\n

These findings reinforce a persistent warning for developers: avoid creating custom SAML implementations. Projects that adopt newer, hardened signature-validation APIs, implement strict element selection, use transform allowlists, impose parser limits, and maintain well-supported identity libraries are better positioned to guard against XML signature confusion and resource-exhaustion attacks.<\/p>\n

Chiang concluded that while AI can assist researchers in systematically testing protocol assumptions at scale, it may also increase the reporting burden on already overextended open-source maintainers.<\/p>\n

Prevent incidents due to slow investigations. Power your Tier 1 with threat intelligence from 15K SOCs:\u00a0Integrate TI\u00a0Lookup in\u00a0your SOC<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"Multiple critical vulnerabilities in SAML implementations after employing Anthropic\u2019s Claude Code in an AI-assisted vulnerability research pipeline. Security…\n","protected":false},"author":2,"featured_media":146103,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[53,3154,182,1393,9644],"class_list":["post-146102","post","type-post","status-publish","format-standard","has-post-thumbnail","category-anthropic","tag-anthropic","tag-anthropic-claude","tag-claude","tag-cyber-security","tag-cyber-security-news"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/146102","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=146102"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/146102\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/146103"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=146102"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=146102"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=146102"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}