This vulnerability in the Strands shell tool allows an attacker to bypass the human consent gate by setting the non_interactive parameter to true directly within the prompt. The shell tool accepted this parameter as truth rather than validating consent through an independent enforcement boundary — essentially placing the permission check inside the trust boundary it was meant to constrain. If your Strands agent is processing untrusted content (scraped pages, documents, retrieved emails), this converts any compromised content source into a potential command injection vector. Upgrade to version 0.8.0 immediately. Before upgrading, audit whether any automation in your deployment pipeline relies on passing the non_interactive parameter — the upgrade path may break existing integrations that depend on this behavior. More critically, determine whether other tools in your Strands deployment (document processing, email, web fetch) use the same consent logic. If the pattern is reused, the same bypass likely applies across the toolchain. Check your dependency trees, container base images, and any pinned versions — versions prior to 0.8.0 will remain exploitable in any environment that doesn't upgrade. The EPSS score is currently low, but this will rise as Strands adoption grows and as the specific injection technique becomes more widely documented in security tooling.
CVE-2026-18733
Official description Straight from the sourceThe vendor's or NVD's own wording, published unedited. Authoritative, but often terse — it says what broke, rarely what to do.
NVD · uneditedA prompt injection vulnerability in the shell tool in Amazon Strands Agents Tools before 0.8.0 might allow remote actors to execute arbitrary operating system commands on the agent's host via a crafted prompt that sets the non_interactive parameter to true, bypassing the human consent gate. To remediate this issue, users should upgrade to version 0.8.0.
Technical summary Written by usOur analysis, written from the advisory, the CVSS vector and the affected-version data. It adds context the advisory leaves out, and never invents facts that are not in the source.
dbcve analysis · high confidenceThis is a prompt injection vulnerability in the shell tool of Amazon Strands Agents Tools. The vulnerability allows remote attackers to bypass the human consent gate by crafting prompts that set the non_interactive parameter to true, enabling execution of arbitrary operating system commands on the host system running the agent.
Verify against the referenced sources before acting — the references below are authoritative for this CVE, this summary is not.
CVSS breakdown How the score is builtThe industry scoring standard. It rates how the flaw is reached, what it takes to exploit, and what an attacker gains — the score is derived from those, not the other way round.
From the vector- Attack vector
- Network
- Complexity
- Low
- Privileges
- None
- User interaction
- Required
- Scope
- Unchanged
- Confidentiality
- High
- Integrity
- High
- Availability
- High
CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
Am I affected? How to checkSteps we derive from the advisory and the affected-version data, so you can decide whether this CVE reaches your setup. They are a guide, not a scan — your own configuration is the authority.
dbcve checksWork through these to decide whether this CVE applies to you.
-
Check if Amazon Strands Agents Tools is installedLook for the strands-agents-tools package in your Python environment using 'pip list' or your package manager. Check for any directory or file named 'strands' or 'amazon-strands' in your project dependencies.Affected if The package is found in your environment
-
Determine the installed versionRun 'pip show strands-agents-tools' or check your lock file (requirements.txt, Pipfile.lock, poetry.lock) for the exact version number of the strands-agents-tools package.Affected if The version is below 0.8.0 or no version is specified (implying an unpatched release)
-
Verify the shell tool is enabledInspect your agent configuration files or code where the Strands Agents Tools are initialized. Look for any tool registration that includes 'shell', 'bash', or 'command' tool definitions.Affected if The shell tool is registered or available in your agent configuration
-
Check for non-interactive parameter handlingReview your agent's prompt templates or tool invocation code to see if the 'non_interactive' parameter can be passed to the shell tool, or inspect the tool's parameter schema.Affected if The shell tool accepts a 'non_interactive' or similar parameter that can be set via prompt injection
You are affected if Amazon Strands Agents Tools is installed with a version before 0.8.0 and the shell tool is enabled in your agent configuration.
Generated from the published advisory. Verify against your own configuration.
Remediation Closing itWhat it takes to close this. Where a vendor fix exists we point at it; where none exists we say so plainly, and can build one. Effort estimates are scoped from the advisory, not from your codebase.
dbcve · scopedUpgrade Amazon Strands Agents Tools to version 0.8.0 or later to remediate the prompt injection vulnerability that allows bypassing the human consent gate.
0.8.0
- Identify the current installed version of Amazon Strands Agents Tools (e.g., run `pip show amazon-strands-agents-tools` or check your requirements file)
- Upgrade the package to version 0.8.0 using pip: `pip install amazon-strands-agents-tools==0.8.0`
- If using a requirements file, update the version constraint to `amazon-strands-agents-tools==0.8.0`
- Verify the upgrade was successful by running `pip show amazon-strands-agents-tools` and confirming the version is 0.8.0
Generated from the published advisory — verify against the referenced sources before acting.
- Consultation2.0 h
- Implementation3.0 h
- Testing3.0 h
- Review / QA2.0 h
An estimate, not a bill — we confirm scope with you before any work starts. Need it this week? Rush from $2,800.
Scan for this in your stack
Free · runs locallyCheck whether your project pulls in CVE-2026-18733 — or any other known-vulnerable package — straight from your lock files. Free and open source; it runs locally and uploads nothing.
References Go to the primary sourcePrimary sources — vendor advisories, patches and trackers. Where our summary and a reference disagree, the reference wins.
Primary sourcesThis vulnerability in the Strands shell tool allows an attacker to bypass the human consent gate by setting the `non_interactive` parameter to `true` directly within the prompt. The shell tool accepted this parameter as truth rather than validating consent through an independent enforcement boundary — essentially placing the permission check inside the trust boundary it was meant to constrain. If your Strands agent is processing untrusted content (scraped pages, documents, retrieved emails), this converts any compromised content source into a potential command injection vector. Upgrade to version 0.8.0 immediately. Before upgrading, audit whether any automation in your deployment pipeline relies on passing the `non_interactive` parameter — the upgrade path may break existing integrations that depend on this behavior. More critically, determine whether other tools in your Strands deployment (document processing, email, web fetch) use the same consent logic. If the pattern is reused, the same bypass likely applies across the toolchain. Check your dependency trees, container base images, and any pinned versions — versions prior to 0.8.0 will remain exploitable in any environment that doesn't upgrade. The EPSS score is currently low, but this will rise as Strands adoption grows and as the specific injection technique becomes more widely documented in security tooling.
Practitioner notes
ContributedPeer-ranked notes from engineers who’ve handled CVE-2026-18733 in production — separate from our analysis above.
The advisory tells you what broke. It rarely tells you what actually worked. If you’ve dealt with this one, that detail is what the next engineer is searching for.
- The version that genuinely resolved it — not the one the vendor claimed
- A config change or rule that shut the vector down
- A gotcha in the upgrade path that cost you an afternoon
A place for practitioners to share what actually worked: a mitigation you’ve tested, a configuration change, a version- or environment-specific caveat, or a link to a verified patch. The most useful notes rise to the top as peers upvote them, so the signal stays high.
- Verified mitigations, workarounds, and config changes
- Version or environment caveats, and links to real fixes
- No weaponised exploit code, or anything meant to cause harm
- No spam, self-promotion, credentials, or personal data