Strands Agents Tools http_request Unauthorized Access Vulnerability CVE-2026-18394 Advisory

CVE-2026-18394 - Unauthorized Access in Strands Agents Tools http_request Tool Vulnerability Overview Strands Agents is an open-source SDK for building AI agents. The package provides pre-built tools, including tools for making API requests via the HTTP request tool. We have identified an unauthorized access issue, CVE-2026-18394, within the tool. Operators can use the allowlist to bind credentials to a set of approved domain names, ensuring they are only sent to those specific hosts. The tool also exposes an parameter in its input mode, which is controllable by Large Language Models (LLMs). Through carefully crafted prompts (e.g., delivered via untrusted web content), an operator can manipulate the agent. The hostname allowlist check still passes on the request URL, allowing credentials to be appended. The request is then routed through the first hop of the proxy, disclosing the credentials in the Authorization header in plaintext. Affected Versions Affected versions: < 0.8.2 Remediation This issue has been resolved in version 0.8.2. We recommend upgrading to the latest version and ensuring that any forked or derived code is patched to incorporate this fix. As a precaution, we advise any operators configuring credentials via on affected versions to rotate these credentials, even if there is no evidence that the credentials have been exposed. Mitigation Before upgrading, do not bind credentials using when the tool is used by agents to process untrusted content. Instead, configure any required out-of-band proxies using the and environment variables rather than through the tool. References CVE-2026-18394 GHSA-g4hw-2h72-m84v Please email aws-security@amazon.com with any security concerns or questions.

Goal Reached Thanks to every supporter — we hit 100%!

Strands Agents Tools http_request Unauthorized Access Vulnerability CVE-2026-18394 Advisory