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-2026-63204 Overview
CVE-2026-63204 is an Insecure Direct Object Reference (IDOR) vulnerability in Zammad, a web-based open source helpdesk and customer support system. Versions prior to 7.1.2 allow an authenticated user with agent permissions to supply an arbitrary AI analytics run identifier to the ticket summarize endpoint. The endpoint returns the AI provider error message stored for that run, even when the run belongs to a ticket the agent cannot access. Disclosure is limited to the provider error string; ticket content is not exposed. The issue is classified under [CWE-639: Authorization Bypass Through User-Controlled Key] and fixed in version 7.1.2.
Critical Impact
Authenticated agents can enumerate AI analytics run error messages for tickets outside their authorization scope through the ticket summarize endpoint.
Affected Products
- Zammad versions prior to 7.1.2
- Zammad ticket summarize controller (app/controllers/ticket/summarize_controller.rb)
- Deployments using the AI analytics summarization feature
Discovery Timeline
- 2026-09-25 - CVE-2026-63204 published to NVD
- 2026-09-28 - Last updated in NVD database
Technical Details for CVE-2026-63204
Vulnerability Analysis
The flaw lives in the Zammad ticket summarize controller. When a client supplies the ai_analytics_run_error_id parameter, the controller previously loaded the referenced AI::Analytics::Run record and returned its stored provider error message directly in the JSON response. No authorization check verified that the requesting agent had access to the associated ticket.
An agent who knows or guesses a valid run identifier can therefore retrieve the provider error string for AI analytics runs on tickets outside their permission scope. The exposure is bounded: only the AI provider error message is returned, not ticket subject, body, or metadata. However, provider error strings can still leak operational details such as model names, prompt fragments echoed back by the provider, or system state useful for further reconnaissance.
Root Cause
The controller performed object lookup by user-supplied identifier without invoking the Pundit policy check (authorize!) required to confirm the caller may view the record. This is a textbook authorization bypass through user-controlled key pattern tracked as [CWE-639].
Attack Vector
Exploitation requires an authenticated session with agent-level permissions and network reachability to the Zammad application. The attacker issues a request to the ticket summarize endpoint with an ai_analytics_run_error_id value referencing a run on a ticket outside their scope and reads the returned result payload. Identifiers are integer-based and may be iterated to enumerate error messages across tenants or groups.
# Security patch in app/controllers/ticket/summarize_controller.rb
# When AI analytics error ID is present, return this error message instead of enqueuing a new job.
if params[:ai_analytics_run_error_id].present?
ai_analytics_run_error = AI::Analytics::Run.find(params[:ai_analytics_run_error_id])
+ authorize!(ai_analytics_run_error, :show?)
render json: {
result: nil,
Source: Zammad GitHub commit 849191c
Detection Methods for CVE-2026-63204
Indicators of Compromise
- Requests to the ticket summarize endpoint containing an ai_analytics_run_error_id parameter from agent accounts that do not own the referenced ticket.
- Sequential or high-volume requests enumerating integer values for ai_analytics_run_error_id from a single agent session.
- Application log entries showing successful AI::Analytics::Run.find calls followed by responses to users without matching ticket group membership.
Detection Strategies
- Correlate web access logs for the ticket summarize route against the authenticated user's ticket group membership to identify cross-scope access attempts.
- Alert on agent accounts generating atypical request volume to AI summarization endpoints or supplying parameters not emitted by the standard Zammad UI.
- Review audit records for AI analytics run reads that do not correspond to a prior legitimate summarize request initiated by the same user.
Monitoring Recommendations
- Forward Zammad application and reverse-proxy logs to a centralized analytics platform and tag requests to AI summarization endpoints for review.
- Establish a baseline for per-agent summarize endpoint usage and alert on deviations exceeding that baseline.
- Monitor for new or compromised agent accounts issuing requests shortly after login to identify automated enumeration behavior.
How to Mitigate CVE-2026-63204
Immediate Actions Required
- Upgrade all Zammad instances to version 7.1.2 or later, which adds the missing authorize! policy check on AI::Analytics::Run lookups.
- Audit agent accounts for recent requests to the ticket summarize endpoint carrying an ai_analytics_run_error_id parameter.
- Rotate credentials or disable accounts showing evidence of enumeration against AI analytics run identifiers.
Patch Information
The fix is included in Zammad 7.1.2. Commit 849191c32a76f62fbd7cf4582a2b1e05e25f25ed adds authorize!(ai_analytics_run_error, :show?) to the ticket summarize controller so the Pundit policy enforces per-ticket authorization before returning stored provider error content. Full details are available in the Zammad GitHub Security Advisory GHSA-75j9-hpg4-j9f8.
Workarounds
- Disable the AI analytics summarization feature until the 7.1.2 upgrade is deployed if operationally feasible.
- Restrict access to the ticket summarize endpoint at the reverse proxy or web application firewall by filtering requests that include the ai_analytics_run_error_id parameter.
- Tighten agent role assignments and review group membership to reduce the number of accounts able to reach the vulnerable endpoint.
# Example reverse-proxy rule to block the vulnerable parameter until patched (nginx)
if ($args ~* "ai_analytics_run_error_id=") {
return 403;
}
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

