CVE-2026-18733 - Prompt injection bypasses shell tool consent gate in Strands Agents Tools
CVE-2026-18733 is a vulnerability in the strands-agents-tools package, specifically in the shell tool that executes OS commands on the agent's host. The shell tool has a human consent gate to approve commands before execution, but a non_interactive parameter exposed in the input schema can be controlled by the large language model (LLM). A crafted prompt, such as one from untrusted content, can set this parameter to true, bypassing the consent gate and allowing arbitrary OS commands to run without operator approval. This affects versions prior to 0.8.0. The issue is fixed in version 0.8.0.
AI Analysis
Technical Summary
Strands Agents is an open-source SDK for AI agents, with strands-agents-tools providing pre-built tools including a shell tool that runs OS commands with operator consent. CVE-2026-18733 arises because the shell tool exposes a non_interactive parameter in its input schema that the LLM can manipulate. An attacker can craft a prompt that sets non_interactive to true, bypassing the human consent gate and enabling arbitrary command execution on the host without operator approval. This vulnerability affects strands-agents-tools versions before 0.8.0. The vendor has addressed the issue in version 0.8.0. Until upgrading, it is recommended not to expose the shell tool to agents processing untrusted content and to run agents in isolated, least-privilege environments.
Potential Impact
An attacker able to deliver crafted prompts to the AI agent can bypass the shell tool's human consent gate and execute arbitrary operating system commands on the host machine without operator approval. This could lead to unauthorized command execution and potential compromise of the host environment. The severity is assessed as low by the vendor, likely reflecting the need for specific conditions such as the ability to influence the agent's input prompts.
Mitigation Recommendations
A fix is available in strands-agents-tools version 0.8.0. Users should upgrade to this version or later to remediate the vulnerability. Until the upgrade can be applied, do not make the shell tool available to agents processing untrusted content. Additionally, run any agent using the shell tool in an isolated, least-privilege environment to contain any executed commands. These mitigations align with the vendor's recommendations.
CVE-2026-18733 - Prompt injection bypasses shell tool consent gate in Strands Agents Tools
Description
CVE-2026-18733 is a vulnerability in the strands-agents-tools package, specifically in the shell tool that executes OS commands on the agent's host. The shell tool has a human consent gate to approve commands before execution, but a non_interactive parameter exposed in the input schema can be controlled by the large language model (LLM). A crafted prompt, such as one from untrusted content, can set this parameter to true, bypassing the consent gate and allowing arbitrary OS commands to run without operator approval. This affects versions prior to 0.8.0. The issue is fixed in version 0.8.0.
Affected software
Run on your own infrastructure? Check whether these packages are installed with threat-finder — our free open-source scanner.
AI-Powered Analysis
Machine-generated threat intelligence
Technical Analysis
Strands Agents is an open-source SDK for AI agents, with strands-agents-tools providing pre-built tools including a shell tool that runs OS commands with operator consent. CVE-2026-18733 arises because the shell tool exposes a non_interactive parameter in its input schema that the LLM can manipulate. An attacker can craft a prompt that sets non_interactive to true, bypassing the human consent gate and enabling arbitrary command execution on the host without operator approval. This vulnerability affects strands-agents-tools versions before 0.8.0. The vendor has addressed the issue in version 0.8.0. Until upgrading, it is recommended not to expose the shell tool to agents processing untrusted content and to run agents in isolated, least-privilege environments.
Potential Impact
An attacker able to deliver crafted prompts to the AI agent can bypass the shell tool's human consent gate and execute arbitrary operating system commands on the host machine without operator approval. This could lead to unauthorized command execution and potential compromise of the host environment. The severity is assessed as low by the vendor, likely reflecting the need for specific conditions such as the ability to influence the agent's input prompts.
Mitigation Recommendations
A fix is available in strands-agents-tools version 0.8.0. Users should upgrade to this version or later to remediate the vulnerability. Until the upgrade can be applied, do not make the shell tool available to agents processing untrusted content. Additionally, run any agent using the shell tool in an isolated, least-privilege environment to contain any executed commands. These mitigations align with the vendor's recommendations.
Technical Details
- Article Source
- {"url":"https://aws.amazon.com/security/security-bulletins/rss/2026-072-aws/","fetched":true,"fetchedAt":"2026-08-03T20:45:32.702Z","wordCount":235}
Threat ID: 6a70fdecbf32cb7a342cfb8e
Added to database: 08/03/2026, 20:45:32 UTC
Last enriched: 08/03/2026, 20:45:42 UTC
Last updated: 08/03/2026, 22:04:16 UTC
Views: 4
Community Reviews
0 reviewsCrowdsource mitigation strategies, share intel context, and vote on the most helpful responses. Sign in to add your voice and help keep defenders ahead.
Want to contribute mitigation steps or threat intel context? Sign in or create an account to join the community discussion.
Actions
Updates to AI analysis require Pro Console access. Upgrade inside Console → Billing.
External Links
Need more coverage?
Upgrade to Pro Console for AI refresh and higher limits.
For incident response and remediation, OffSeq services can help resolve threats faster.
Latest Threats
Check if your credentials are on the dark web
Instant breach scanning across billions of leaked records. Free tier available.