AmazonAnalysisAWS fixes Strands Agents Tools secret masking bypass in environment tool
What happened.Amazon Web Services published security bulletin AWS-2026-154 disclosing a high-severity secret masking bypass vulnerability in the open-source Strands Agents Tools developer package.
Why it matters.Autonomous AI agents require granular capability limits when interacting with host execution environments. When tool definitions permit language models to disable defensive controls, untrusted input can siphon long-lived cloud credentials directly. Engineering teams running AI agent workflows must upgrade dependencies to version 0.8.6 and audit prompt injection defenses. Platform architects should strip permanent cloud credentials from agent runtime execution processes.
The technical specifications:
- Vulnerability mechanics: Tracked as CVE-2026-115415 with an 8.6 CVSS score, the environment tool in strands-agents-tools exposed its secret masking toggle directly within the model tool schema.
- Exploitation path: Malicious or untrusted prompts processed by autonomous agents could instruct the model to disable masking, dumping host environment variables and cloud API tokens in cleartext.
- Remediation releases: AWS published strands-agents-tools version 0.8.6, deprecating the environment tool, enforcing strict operator controls over masking parameters, and instructing developers to rotate exposed API credentials.
STORYLINE PROGRESSION & CONTEXT
This report represents a verified evolution of prior newsroom intelligence: