External risk intelligence

Flowise CSV Agent Prompt Injection Executes Arbitrary Code

CVE advisorySeverity: CRITICAL (CVSS 9.5)

CVE-2026-70477

Flowise is a tool used to build and deploy large language model interfaces. These flows are typically exposed as web applications or API endpoints to allow user interaction, making the interface and its underlying processing nodes, such as the affected CSV Agent, commonly reachable via the public internet in standard deployment patterns.

Code Injection

Halo Surface Signal: 4 out of 5 — likely to be public-facing.

External exposure likelihood

Horizon Alert

Summary of the vulnerability and why it matters

A vulnerability exists in Flowise, a tool for building custom large language model interfaces, where a specially crafted prompt can lead to the execution of malicious Python code. This could allow an attacker to run arbitrary commands within the service's environment. The main concern is confirming if your environment uses the affected technology and is exposed.

  • Prompt injection allows code execution.
  • Critical to verify if your LLM interface is affected.
  • Confirm exposure and relevance for your LLM deployments.

Attack Path

How an attacker could exploit the issue

An attacker can exploit this vulnerability by sending a specially crafted prompt to a chatflow that uses a CSV Agent node. If the prompt is accepted, the underlying large language model may generate a malicious Python script. This script can bypass security checks and execute arbitrary code within the application's environment.

  • Attacker sends malicious prompt.
  • CSV Agent node processes prompt.
  • Arbitrary code execution occurs.

Live Threat

Current exploitation, exposure, and threat context

When supported by the advisory, an attacker could send a malicious prompt to a chatflow using a CSV Agent node. This could cause the LLM to generate and execute a Python script, bypassing security checks and running arbitrary code within the service's context.

  • Arbitrary code execution in service account.
  • Malicious script bypasses validation.
  • Compromise of service and its data.

Operational Fix

Recommended remediation, mitigation, and detection steps

Ownership of this vulnerability likely resides with the team managing the Flowise application and its underlying LLM integrations. The first critical step is to identify all instances of Flowise, determine their exposure and business criticality, and then confirm the accountable owner for each. This will inform a prioritized remediation plan.

  • Assign ownership to Flowise application managers.
  • Verify Flowise deployment exposure and criticality.
  • Plan and coordinate vendor-supported updates.

Supplementary metadata

Validate whether this threat affects your internet-facing exposure.

Halo Threat Intelligence helps prioritize remediation with Halo Surface Signal and H/A/L/O context. Start exposure validation with a free external attack surface trial.

Frequently asked questions

What is Flowise?

Flowise is a developer tool that provides a visual drag-and-drop interface for creating and managing customized workflows for large language models. It allows users to connect various nodes to build intelligent chat interfaces and data-processing pipelines. A key component of these workflows is the CSV Agent node, which is designed to help the system interact with and interpret CSV data files during a chat session.

What is the security weakness in CVE-2026-70477?

This vulnerability is classified as CWE-94: Improper Control of Generation of Code. In simple terms, it means the application inadvertently allows external input to influence the creation and execution of code. Specifically, in Flowise, a maliciously crafted prompt can trick an LLM into generating unauthorized Python code. This generated script then exploits a flaw in how the system validates code, allowing it to run dangerous commands that should have been blocked.

How is this vulnerability triggered?

An attacker triggers this by interacting with a chatflow that utilizes a CSV Agent node. The attack succeeds when the attacker sends a prompt designed to manipulate the LLM into producing a malicious Python script. It is important to note that this specific bug is tied to the use of the CSV Agent node; chatflows or applications that do not utilize this specific node configuration do not encounter this particular execution path.

Is my Flowise deployment at risk?

According to Halo Surface Signal, Flowise instances are often deployed as public-facing web applications or API endpoints to facilitate user interaction. If your instance is reachable via the public internet, it faces a higher level of risk because attackers can reach the CSV Agent node directly. You should assess whether your specific deployments are internet-accessible or restricted to internal users to gauge your immediate threat level.

How do I respond to this threat?

The most effective first step is to verify the version of Flowise currently running in your environment. If you are using any version prior to 3.1.3, you should prioritize upgrading to version 3.1.3 or later, which contains the fix for this issue. Coordinate with the team managing your application infrastructure to identify all active instances and schedule the necessary updates to ensure the vulnerability is patched.

References