Secret masking bypass vulnerability in the environment tool of Strands Agents Tools

Strands Agents Tools released a security advisory (GHSA-qjx4-hg6g-2f34 / CVE-2026-115415): the secret masking of its environment tool can be bypassed, and this has been fixed in version 0.8.6. The tool masks environment variables whose names contain KEY, TOKEN, SECRET or PASSWORD by default, but the masked parameter that controls masking is written into the tool input schema, so the model can set it itself, and the list and get operations require no operator confirmation. An attacker able to inject instructions into content processed by the Agent (e.g., via web pages, files or prior tool output) can induce the model to call the tool with action="list", masked=false, returning in plaintext AWS access keys, model provider API keys, source control tokens and database connection strings; the model may then pass these values to the attacker through other loaded tools.