CVE-2026-82270

Portkey AI Gateway through version 1.15.2 contains a server-side request forgery flaw (CWE-918) in its /v1/proxy/* route, which is exposed without the requestValidator middleware. An unauthenticated remote attacker can send a request to this proxy route with the x-portkey-custom-host header set to an internal address, causing the gateway to forward the request to attacker-chosen internal hosts. Because the gateway forwards these requests with its configured Authorization headers, the attacker can reach internal services and exfiltrate provider API keys, which drives the high confidentiality impact behind the 8.7 (High) CVSS 4.0 score. Any deployment of the open-source, self-hosted Portkey AI Gateway at version 1.15.2 or earlier is affected, particularly where the proxy route is reachable by untrusted or internet-facing clients. No public proof of concept, CISA KEV listing, or confirmed in-the-wild exploitation is known, and EPSS assigns a low 0.3% probability of exploitation within 30 days (20th percentile).

What to do: Upgrade to a Portkey AI Gateway release newer than 1.15.2 as soon as a patched version is published (no fixed version is specified in the available data). Until then, restrict or protect the /v1/proxy/* route — for example by adding the requestValidator middleware or blocking/validating the x-portkey-custom-host header at a reverse proxy — prevent proxying to internal or private address ranges, review logs for proxy requests that set x-portkey-custom-host, and rotate provider API keys for any internet-exposed deployment.