Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-94613 Overview
CVE-2026-94613 is a denial-of-service vulnerability in authentik, an open-source identity provider. An unauthenticated attacker can submit a malformed Security Assertion Markup Language (SAML) message to any authentik deployment that uses SAML as either an identity provider or a SAML source. The malformed message terminates the worker handling /application/saml/* or /source/saml/* routes, causing in-flight requests on that worker to fail. Although workers restart automatically and database-backed sessions survive, continued malicious traffic sustains the outage against a share of legitimate SAML traffic. The flaw is fixed in authentik 2026.2.7, 2026.5.7, and 2026.8.2.
Critical Impact
Unauthenticated remote attackers can sustain partial denial of service against SAML-based authentication flows, disrupting single sign-on for federated users.
Affected Products
- authentik versions prior to 2026.2.7
- authentik versions prior to 2026.5.7
- authentik versions prior to 2026.8.2
Discovery Timeline
- 2026-09-24 - CVE-2026-94613 published to NVD
- 2026-09-29 - Last updated in NVD database
Technical Details for CVE-2026-94613
Vulnerability Analysis
The flaw is an uncontrolled resource consumption weakness [CWE-770] in authentik's SAML message handling. When the SAML provider or SAML source endpoint parses an attacker-supplied XML message, the parser raises an unhandled condition that crashes the worker process. The crash terminates every request currently assigned to that worker, including requests from unrelated users and tenants.
Automatic worker restart restores capacity, yet the restart window remains an availability gap. An attacker can send malformed messages continuously to keep a measurable fraction of SAML traffic failing. Other authentik protocol implementations such as OAuth2 and LDAP are not affected. Database-backed sessions persist across restarts, so already-authenticated users retain their sessions.
Root Cause
The root cause is unsafe XML parsing in authentik/lib/xml.py used by authentik/providers/saml/processors/authn_request_parser.py. The parser accepted document type declarations and other XML features that could trigger fatal parser conditions. The fix introduces an UnsafeXML exception and a _reject_doctype helper that pre-scans input with expat before handing it to lxml, rejecting documents containing a DOCTYPE declaration.
Attack Vector
Exploitation requires only network access to a SAML endpoint. The attacker submits a crafted SAML request or response to /application/saml/<slug>/ or /source/saml/<slug>/. No credentials, user interaction, or prior enrollment are required.
# Patch excerpt from authentik/lib/xml.py introducing DOCTYPE rejection
"""XML Utilities"""
from typing import NoReturn
from xml.parsers import expat # nosec
from lxml.etree import XMLParser, _Element, fromstring, tostring # nosec
class UnsafeXML(ValueError):
"""A document uses XML features that are not accepted."""
class _RootElement(Exception):
"""The scan reached the root element. expat offers no other way to stop it."""
class _Doctype(Exception):
"""The scan reached a document type declaration."""
def _reject_doctype(data: bytes) -> None:
"""Refuse a document type declaration.
The check runs on expat rather than on the parser below, and looks for a
declaration rather than for its text, because a document may be in any encoding.
Parsing stops at the root element, so only the prolog is read."""
parser = expat.ParserCreate()
Source: Authentik Commit #03d19d63
Detection Methods for CVE-2026-94613
Indicators of Compromise
- Repeated worker process restarts logged by the authentik worker container or systemd unit
- Elevated HTTP 5xx responses on URIs matching /application/saml/* and /source/saml/*
- SAML request bodies containing <!DOCTYPE declarations or other prolog-level XML directives
- Short bursts of failed SAML authentication attempts from a single source address
Detection Strategies
- Alert on authentik worker restart counts exceeding baseline over a rolling 5-minute window
- Inspect reverse proxy and ingress logs for SAML endpoint requests returning HTTP 500 or 502
- Flag inbound SAML payloads whose decoded, inflated XML contains a DOCTYPE node
- Correlate SAML endpoint failures with concurrent success on OAuth2 or LDAP flows to confirm protocol-specific disruption
Monitoring Recommendations
- Export authentik worker metrics (Prometheus authentik_outpost_* and process liveness) to a centralized monitoring system
- Enable WAF or API gateway logging of request bodies on SAML endpoints for forensic review
- Monitor availability of SAML single sign-on flows with synthetic transactions from external probes
How to Mitigate CVE-2026-94613
Immediate Actions Required
- Upgrade authentik to 2026.2.7, 2026.5.7, or 2026.8.2 depending on your current release train
- Restrict access to /application/saml/* and /source/saml/* from untrusted networks until patched
- Deploy WAF rules that reject SAML POST bodies containing <!DOCTYPE after base64 decode and inflate
Patch Information
The maintainers released fixes in authentik 2026.2.7, 2026.5.7, and 2026.8.2. The code changes land in commits 03d19d63, 4cf2f803, 732adad2, and a029f537. Full disclosure is published in GHSA-cxwx-9x59-28qm.
Workarounds
- Scale authentik worker replicas horizontally so a single crashed worker handles a smaller share of SAML traffic
- Place rate limits on SAML endpoints at the ingress or reverse proxy layer to slow repeated malformed submissions
- Temporarily disable SAML sources or SAML providers that are not business-critical until the upgrade is applied
# Example upgrade for a Docker Compose deployment
export AUTHENTIK_TAG=2026.8.2
docker compose pull
docker compose up -d
# Example upgrade for a Helm-based Kubernetes deployment
helm repo update
helm upgrade authentik authentik/authentik \
--namespace authentik \
--set image.tag=2026.8.2
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

