Envoy CVE-2026-6994

Exploitation requires authenticated access to Envoy's configuration APIs or proxy administration interface (PR:L per CVSS vector) to inject malicious query parameter rules into the Header Mutation filter configuration. The Header Mutation filter itself must be enabled and configured to process query parameters using the params.add function. Default Envoy installations without explicit Header Mutation filter configuration are not vulnerable; vulnerability only manifests when an administrator has intentionally configured this filter for query parameter manipulation. Attackers cannot exploit this remotely without valid credentials to modify proxy configurations.

The CVSS 4.0 vector (AV:N/AC:L/PR:L/UI:N) indicates network-accessible exploitation requiring low-complexity authenticated access with no user interaction, affecting confidentiality and integrity at low impact levels. The low CVSS score of 5.3 reflects the authentication requirement (PR:L) and limited impact scope rather than underestimating severity - this is a legitimate medium-risk vulnerability rather than critical. However, the placement in a core HTTP filtering component that processes all proxied traffic, combined with code injection classification, suggests real-world risk exceeds the numerical score: successful exploitation in a multi-tenant proxy environment could allow lateral movement or privilege escalation beyond the CVSS scope boundary. No KEV listing or public exploit code has been identified, but the upstream fix commitment (referenced commit hash) and specific file/function targeting indicate vendor confidence in the issue's reproducibility. The authentication barrier is significant - exploitation requires an account with proxy configuration privileges, reducing opportunistic attack surface but not eliminating insider risk.

An authenticated Envoy administrator or operator with proxy configuration privileges modifies or injects a malicious query parameter rule into the Header Mutation filter configuration. When Envoy processes HTTP requests matching the configured rule, the injected query parameter value breaks out of the intended parameter context and executes code within the filter's processing chain, potentially extracting or modifying request/response data before forwarding to upstream services or clients. The low attack complexity and lack of required user interaction mean exploitation can be automated once initial access to configuration APIs is gained.