Users providing user generated input into the resolveEndpoint method on requests.
Upgrade to Saloon v4+
Upgrade guide: https://docs.saloon.dev/upgrade/upgrading-from-v3-to-v4
When building the request URL, Saloon combined the connector's base URL with the request endpoint. If the endpoint was a valid absolute URL (e.g. https://attacker.example.com/callback), the code used that URL as-is and ignored the base URL. The request—and any authentication headers, cookies, or tokens attached by the connector—was then sent to the attacker-controlled host. If the endpoint could be influenced by user input or configuration (e.g. redirect_uri, callback URL), this allowed server-side request forgery (SSRF) and/or credential leakage to a third-party host. The fix (in the next major version) is to reject absolute URLs in the endpoint: URLHelper::join() throws InvalidArgumentException when the endpoint is a valid absolute URL, unless explicitly allowed, requiring callers to opt-in to the functionality on a per-connector or per-request basis.
Saloon thanks @HuajiHD for finding the issue and recommending solutions and @JonPurvis for applying the fix.
{
"cwe_ids": [
"CWE-522",
"CWE-918"
],
"nvd_published_at": "2026-03-26T01:16:27Z",
"severity": "MODERATE",
"github_reviewed": true,
"github_reviewed_at": "2026-03-25T22:00:13Z"
}