Incorrect authorization in the http_request tool in Strands Agents Tools before 0.8.2 might allow remote attackers to obtain credentials configured via HTTP_REQ…

CVE-2026-18394 is a vulnerability in the http_request tool of Strands Agents Tools prior to version 0.8.2. The flaw involves incorrect authorization logic that can be exploited by remote attackers to obtain credentials stored in HTTP_REQUEST_TOKEN_CONFIG. The attack vector involves influencing the underlying LLM to route HTTP requests through attacker-controlled proxy infrastructure, effectively leaking sensitive authentication tokens. This represents a prompt injection or indirect manipulation attack targeting AI agent tooling. The vulnerability has been assigned a high severity rating. AWS has published a security bulletin regarding this issue. The fix is available in version 0.8.2 of the Strands Agents Tools package. Users are strongly advised to upgrade immediately to mitigate credential theft risks.