AWS discloses a Strands tool secret-masking bypass

AWS disclosed a Strands Agents Tools defect that let model-controlled tool inputs disable environment-variable masking. Crafted content could expose credentials without operator consent; version 0.8.6 made masking operator-controlled and deprecated the environment tool.

What the public record establishes

AWS and the Strands maintainers published their advisories on October 28. They identify affected strands-agents-tools versions from 0.1.0 through versions below 0.8.6. The date records public disclosure; neither advisory establishes that the fix first became available that day. [1][2]

The model could set a masking parameter in the environment tool’s schema. Instructions in processed content could therefore cause cleartext environment-variable output without an operator consent prompt. The advisories identify 0.8.6 as patched and deprecate the tool; they do not report malicious exploitation. [1][2]

Sources

  1. CVE-2026-115415 - Secret masking bypass in the environment tool in Strands Agents Tools

    Amazon Web Services · Oct 28, 2026

  2. Secret masking bypass in the environment tool in Strands Agents Tools

    Strands Agents · Oct 28, 2026

  3. Strands Agents Tools: a prompt turns off the environment tool's masking

    The Injection · Oct 28, 2026