CVE-2026-133631: FlowIntel chatbot let LLM call any API, exposed Ollama token in process table

FlowIntel's AI chatbot allowed language models to call any API endpoint without validation, and AI report generation exposed the Ollama bearer token in the system process table. Fixed in commit 34d151b, shipping in version 4.1.0.

Christopher Ort
By i10X News Desk · Reviewed by Christopher Ort · 4 min read
Share
A security analyst reviews case files on a laptop in a dimly lit operations center, with redacted documents and threat intelligence reports visible on surrounding monitors.
Image: AI-generated, i10X
On this page
  1. Quick Take
  2. In plain English
  3. What happened
  4. What you should do
  5. What it means
  6. Sources

FlowIntel, an open-source case management platform for security and threat intelligence analysts, contained three security flaws in its AI chatbot and report generation features. The chatbot API tool permitted the language model to invoke any FlowIntel API endpoint without an allow-list or path validation, and AI report generation passed the Ollama bearer token and full case content as command-line arguments visible to any local user.

i10X Quick Take

Quick Take

  1. What: CVE-2026-133631, three vulnerabilities in FlowIntel’s AI chatbot and report generation (CVSS 7.1 high)
  2. Who is affected: FlowIntel installations prior to 4.1.0 using the AI chatbot or Ollama-based report generation
  3. Chatbot risk: Language model could invoke administrative API routes and state-changing GET operations without validation
  4. Token exposure: AI report requests passed Ollama bearer token and case content as command-line arguments visible in the system process table
  5. What to do: Update to commit 34d151b on the main branch; version 4.1.0 will include the fix when released

In plain English

FlowIntel is a tool that security analysts use to manage investigations and generate reports with AI help. A security researcher reported that the AI chatbot feature could be tricked into calling any internal function, including admin commands, if an attacker could influence what the AI reads (for example, by adding malicious text to a case). Separately, when the system generated AI reports using Ollama, it accidentally showed the secret access token and case details in a place where any user on the same computer could see them. The developers have fixed these problems in a code update.

What you should do: If you use FlowIntel with AI features, you should update to the latest code from the main branch. If you don’t use FlowIntel, you can ignore this.

What happened

The National Vulnerability Database published CVE-2026-133631 on November 20, 2026, describing three security flaws in FlowIntel prior to version 4.1.0. FlowIntel is an open-source case management platform for security and threat intelligence analysts with an optional AI chatbot and Ollama-based AI report generation.

The first flaw involved note variable references. When a user embedded references to other cases or tasks in a note, FlowIntel resolved those references using the viewing user’s privileges rather than checking whether the destination case’s audience had granted read access. An authenticated user with access to a shared or imported note could embed references to private cases, causing sensitive case data to be pulled into AI report requests, exports, and MISP sends without proper authorization.

The second vulnerability affected the chatbot API tool. The tool permitted the language model to invoke any FlowIntel API endpoint, including administrative routes and state-changing GET operations, without an allow-list or path validation. An attacker who could influence the model’s instructions, for example via untrusted case content, could induce data modification or access restricted endpoints.

The third issue involved AI report generation. The system passed the Ollama bearer token and full case content as command-line arguments to a child process, exposing them in the system process table to any local user on the same machine.

The NVD entry assigns a CVSS 4.0 score of 7.1 (high severity) and lists CWE-200 (Exposure of Sensitive Information), CWE-284 (Improper Access Control), and CWE-862 (Missing Authorization).

What you should do

The FlowIntel maintainers published a fix in commit 34d151b on November 3, 2026. The commit message credits the report to a researcher using the handle “elhoim”.

Note variable references now require the destination case’s audience to be allowed to read the referenced case. The chatbot API tool now defaults to an explicit allow-list of read operations and searches, including protection against state-changing GET endpoints. Administrative endpoints and ambiguous or external paths are refused. Validated paths match the pinned MCP transport, and other operations require an explicit CHATBOT_ALLOW_WRITES opt-in.

AI report requests now reach the child process through private anonymous temporary stdin instead of command-line arguments, keeping the Ollama bearer token and case content out of argv.

As of November 20, 2026, the latest tagged GitHub release is 4.0.0. The fix is available in commit 34d151b on the main branch and will ship in version 4.1.0 when that release is published.

What it means

The chatbot vulnerability is a form of prompt injection: an attacker who can place malicious instructions in case content that the AI reads could trick the model into calling administrative API endpoints or modifying data. The token exposure in the process table is a classic information disclosure flaw that could allow any local user to steal credentials and case data.

No exploitation in the wild is reported in the public sources. Organizations running FlowIntel with AI features enabled should update to the fixed commit to prevent unauthorized API access and credential leakage.

Language models that can call tools beyond their intended scope are a recurring pattern, as in the three MCP servers that let AI agents break out of tool limits, and leaked model credentials echo the LLM 0.37.2 API key leak. New cases are tracked in our AI Security hub.

Sources

  • CVE-2026-133631: NVD
  • Fix commit 34d151b: flowintel/flowintel GitHub
Christopher Ort

Christopher Ort

Christopher Ort is the editor of i10X News at i10x.ai. He has worked hands-on with AI tools, models and agent workflows for years and leads i10X’s coverage of AI launches, funding and research. He is the author of “The Style Penalty”, an SSRN study on how frontier AI models judge AI-written CVs, covered by heise online and t3n.

Editor's picks

  1. Dragos canary study: attackers stole AI prompts and credentials at three AI inference providers
  2. Analysis: AI Agents in Banking Hit a Liability Wall
  3. CVE-2026-133631: FlowIntel chatbot let LLM call any API, exposed Ollama token in process table
  4. Stripe Evals: Coding Agents Score 78% on Payment Integrations, but Only 29% Pass Every Required Check
  5. Kong Volcano: A Self-Healing Agent Cloud Where Coding Agents Build, Host and Fix Apps