Dgraph patches pre-auth DQL injection via upsert cond

TL;DR - Dgraph < 25.3.3 can be queried pre-auth: one POST /mutate?commitNow=true with a crafted upsert cond injects DQL and returns arbitrary query results.

What happened

Dgraph is an open source distributed GraphQL database with an HTTP API that supports mutations and upserts.

CVE-2026-41327 is a pre-auth data-exfiltration flaw when ACL is not enabled (the default configuration called out in the record). The attack is a single HTTP POST to /mutate?commitNow=true that supplies a crafted cond field in an upsert mutation.

The core issue is that the cond value is concatenated directly into a DQL query string via strings.Builder.WriteString after only a cosmetic strings.Replace transformation. There is no escaping, parameterization, or structural validation. An attacker can inject an additional DQL query block into cond; the injected query executes server-side and its results are returned in the HTTP response.

This is high-leverage because it turns a commonly exposed application-facing database API into a full read primitive under default no-ACL deployments.

Who is impacted

• Dgraph (dgraph-io/dgraph) versions < 25.3.3.
• Environments running Dgraph with ACL not enabled (called out as the default configuration in the record), where the HTTP API is reachable by untrusted clients.

What to do now

• Follow vendor remediation guidance and apply a release that includes the fix.

  "This vulnerability is fixed in 25.3.3."

• Inventory Dgraph deployments and confirm whether Alpha HTTP endpoints (notably /mutate) are reachable from the internet or broad internal networks.
• Treat this as potential data exposure if you were running without ACL and had network exposure:
  • review request logs for unexpected POST traffic to /mutate?commitNow=true
  • specifically flag requests containing suspicious or unusually structured cond values
• Reduce blast radius while patching where operationally feasible:
  • restrict network access to Dgraph HTTP ports to only trusted callers
  • avoid relying on default no-ACL posture for any Dgraph instance holding sensitive data

Additional Information

CISA ADP vulnrichment for this CVE marks Exploitation: poc and Automatable: yes.