CVE-2026-110782: SAML2: Incomplete fix of CVE-2026-49283: unsigned embedded Response bypasses HTTP-Artifact signature verification (no embedded-Issuer binding)

High CVSS 8.7

OSV id
GHSA-r7hw-jx6r-756g
Also known as
CVE-2026-110782
Published
Last modified

Affected packages

PackageAffected versionsFixed in
simplesamlphp/saml2
Packagist
>= 6.2.2, < 6.2.4 6.2.4
simplesamlphp/saml2
Packagist
>= 7.0.0-rc1, < 7.0.0-rc2 7.0.0-rc2

How to fix it

simplesamlphp/saml2 (Packagist)

Upgrade simplesamlphp/saml2 to 6.2.4 or later.

composer update simplesamlphp/saml2
# if composer.json blocks it: composer require "simplesamlphp/saml2:^6.2.4"
  • If it is an indirect dependency, `composer update` the package that requires it, with dependencies.
  • Check the changelog for breaking changes (especially across major versions), update the lockfile with the package manager, and run your tests.

simplesamlphp/saml2 (Packagist)

Upgrade simplesamlphp/saml2 to 7.0.0-rc2 or later.

composer update simplesamlphp/saml2
# if composer.json blocks it: composer require "simplesamlphp/saml2:^7.0.0-rc2"
  • If it is an indirect dependency, `composer update` the package that requires it, with dependencies.
  • Check the changelog for breaking changes (especially across major versions), update the lockfile with the package manager, and run your tests.

Not sure whether your project uses an affected version? Check your lockfile below for free.

Details

Summary

CVE-2026-49283 / GHSA-6929-8p9f-26jx ("HTTP-Artifact TLS validator confusion allows cross-IdP authentication bypass") was fixed so that HTTPArtifact::receive() verifies the embedded SAML Response using the IdP metadata signing keys. The current implementation (master 31c72c1d, released v6.2.3) reaches that verification only when the embedded Response is signed:

// src/Binding/HTTPArtifact.php  (receive())
if (!$samlResponse->isSigned()) {
    return $samlResponse;                         // ← unsigned embedded Response returned UNVERIFIED
}
return $this->verifyMessageSignature($samlResponse, $idpMetadata);   // only runs when signed

An attacker who controls one trusted-federation IdP therefore returns a validly-signed outer ArtifactResponse (signed with their own real key, which passes the mandatory outer check) that wraps an unsigned <samlp:Response> whose <Issuer> names a different IdP. isSigned() is false, so verifyMessageSignature() never runs and the forged Response is returned as successfully received. Additionally, even on the signed path verifyMessageSignature() verifies against the artifact issuer's metadata keys but never asserts that the embedded Response's <Issuer> equals the artifact issuer — the same cross-IdP confusion the CVE was assigned for is not closed by identity binding.

This is an incomplete implementation of the fix's own stated goal ("verify signatures using IdP metadata signing keys", commit dac0b4b6 / #419), and a regression introduced by the #414 refactor (v6.2.2) — the official CVE fix in v6.2.1 did not contain this short-circuit.

Affected component

  • src/Binding/HTTPArtifact.php
    • receive() line 206-207: if (!$samlResponse->isSigned()) { return $samlResponse; } — the bypassed guard.
    • receive() line 210: verifyMessageSignature($samlResponse, $idpMetadata) — only reached when signed.
    • verifyMessageSignature() lines 230-286: verifies against $idpMetadata (artifact-issuer keys); no getIssuer() comparison to bind the embedded Response to the artifact issuer.
    • verifyArtifactResponseSignature() line 300-302: if ($artifactResponse->isSigned() !== true) throw — the OUTER response is mandatorily signed; the asymmetry with the embedded return-on-unsigned is the defect.
  • Regression commit 3b589636 (#414, v6.2.2, "refactor/soap-ssl-validator-openssl"); key-source fix dac0b4b6 (#419) left the short-circuit unchanged.
  • Affected: v6.2.2, v6.2.3 (latest), and master. v6.2.1 (official CVE fix) used a deferred-validator model without this short-circuit.

Reachability (hop-by-hop, guards named)

Threat model: SP uses the HTTP-Artifact binding in a multi-IdP federation; attacker operates one trusted IdP-B (valid saml20-idp-remote metadata, real signing key, an ArtifactResolutionService). Goal: impersonate admin@IdP-A.

  1. Attacker hits the SP artifact endpoint with SAMLart encoding sourceId = sha1(IdP-B entityID) — line 137-138, no guard.
  2. getMetaDataConfigForSha1 → $idpMetadata = IdP-B — line 144. Guard: metadata exists → satisfied (IdP-B trusted).
  3. SP resolves against IdP-B's ARS (attacker's endpoint) — line 176-182.
  4. IdP-B returns a validly-signed ArtifactResponse; verifyArtifactResponseSignature(...,IdP-B) passes (signed by the artifact issuer) — line 193/300. Guard satisfied by the attacker (they ARE the artifact issuer).
  5. The signed outer wraps an unsigned <samlp:Response Issuer=IdP-A> with an assertion for admin@IdP-A — line 195, no guard on embedded content.
  6. if (!$samlResponse->isSigned()) return $samlResponse; → TRUE → returns the forged Response unverified — line 206-207. verifyMessageSignature (210) never runs; there is no issuer==artifact-issuer check anywhere in receive().

Every guard on the embedded message is absent; the outer/embedded mandatory-vs-optional asymmetry is the root cause.

Root cause

The #414 refactor moved embedded-Response verification into receive() but gated it on isSigned() (mirroring if (!isSigned()) return instead of the outer's if (isSigned() !== true) throw), and the dac0b4b6 key-source fix did not revisit that gate. The artifact binding's designed trust model is that the embedded message is authenticated by the binding (the assertion need not be individually signed), so a conformant integrator relies on receive() to have authenticated it — exactly the reliance this gap defeats.

Proof of concept (executed, benign)

The PoC exercises the real shipped private methods verifyArtifactResponseSignature() and verifyMessageSignature() via reflection — exactly as the maintainers' own tests/SAML2/Binding/HTTPArtifactTest.php does — and reproduces receive()'s verbatim post-resolution decision (lines 195/206-210) against real HTTPArtifact internals. It models a validly-signed outer ArtifactResponse from the artifact issuer IdP-B (attacker

References

Am I affected?

Upload or paste your lockfile. We check every package version in it against OSV.dev, this advisory included, and email you the vulnerable ones with the version that fixes them. Free, and no account needed.

or paste its content

Needed when you paste. package.json and pyproject.toml only have version ranges: use the lockfile next to them.

Only package names and versions are checked; the file itself is not kept. We use your email to send this report and nothing else unless you tick the box above. Privacy policy.

About this data

Source: GitHub Advisory Database via OSV.dev (CC-BY 4.0). Severity comes from the advisory's own rating when it has one, else from its CVSS score.