Docker Sandboxes may forward a client‑supplied authentication header together with a credential injected by the host egress proxy. The proxy removes alternate credentials only when their exact values match known sentinel values. Consequently, code running inside an authorized sandbox can supply an unrecognised credential in a supported authentication header. That unrecognised credential is forwarded to the upstream service, potentially authenticating to an attacker‑controlled account and exposing data that was included in the original request. The weakness is documented as CWE‑200, representing a confidentiality exposure.

The vulnerability affects Docker Sandboxes. All versions released prior to Docker Sandboxes 0.43.0 are vulnerable; upgrading to 0.43.0 or later resolves the issue. No specific minor releases are listed.

The CVSS base score of 5.9 indicates moderate severity, and the EPSS score is not available, giving no direct measure of current exploitation likelihood. The vulnerability does not appear in CISA's KEV catalog. The likely attack vector requires an attacker to execute code within an authorized sandbox that has permission to inject custom credentials. If an attacker can run arbitrary code in that sandbox, they could manipulate outward requests to authenticate as an unintended account on downstream services, resulting in data exposure. The scope of the impact is limited to downstream services that accept such credentials and the data they process.