Author: Nadali. Prepared using AI.
On October 28, AWS published a bulletin about CVE-2026-115415 in the Python package strands-agents-tools. Versions from 0.1.0 inclusive to 0.8.6 exclusive are affected; 0.8.6 contains the fix.
What changed
According to the official advisory, the environment tool could expose secrets from an agent’s environment through instructions injected into the model’s input. After the fix, the model can no longer disable masking; the tool itself is deprecated.
This concerns businesses whose own or contracted AI agents use this specific package and tool. It does not mean every AWS or chatbot user has this vulnerability.
What to ask your contractor
Our editorial review sequence:
- Request the installed package version and confirmation of whether your agents loaded environment.
- If they did, agree on an upgrade and review of modified copies of the code. AWS also recommends rotating credentials that were present in those agents’ environments.
- If an immediate upgrade is impossible, the provider recommends removing environment from the agents. Agree on a manual fallback for tasks that depend on it.
- Obtain confirmation of the version after the upgrade and checks of your workflows. Do not send the actual keys in correspondence for “verification.”
Limits and takeaway
The notices we read do not claim confirmed active exploitation. Nadali did not test the vulnerability or inspect readers’ systems.
The practical takeaway: first establish whether the notice applies to your automation, then agree on remediation and credential handling with the responsible specialist.
Related: why incoming documents must not become AI instructions. Sources read on October 31, 2026. Cover: a reused AI illustration, not an image of the incident.
