What happened
CVE-2026-115415 (CVSS 6.5): improper neutralization of input used for LLM prompting in the environment tool of Strands Agents Tools before 0.8.6. A crafted prompt can disable the tool's secret masking, letting a remote actor obtain the agent host's environment variables in cleartext, including cloud credentials and API keys. AWS published bulletin 2026-154-AWS and the fix is in 0.8.6.
Why it matters
An agent that dumps environment variables is a direct credential-disclosure path: a single crafted prompt turns the environment tool into a source of cloud API keys and secrets, which is a core AI-agent risk. AWS issued the advisory, so the affected tooling is in the agent-ecosystem mainstream.
Attack vector
A remote actor crafts a prompt that disables the environment tool's secret masking so the agent returns the host environment variables (cloud credentials, API keys) in cleartext.
Affected systems
Strands Agents Tools 0.1.0 to before 0.8.6.
Mitigation
Upgrade to Strands Agents Tools 0.8.6 or later; rotate any credentials present in the environment of agents that loaded the environment tool; avoid placing secrets in agent process env vars.