# CVE-2026-105135: InternLM MindSearch code injection in planner agent

Source: https://junglewise.ai/threats/cve-2026-105135-a-vulnerability-has-been-found-in-internlm-mindsearch-0-1-0-this  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-10-04. Updated: 2026-10-05  
License: CC BY 4.0 with a link back

CVE-2026-105135 · Severity: critical · CVSS 10 · Published 2026-10-04

InternLM MindSearch is an AI agent framework that processes web search queries through a planning component. An unauthenticated attacker can send a crafted request to the exposed API endpoint and inject malicious Python code that executes with full server privileges, allowing remote code execution. The vulnerability requires no authentication, credentials, or user interaction, putting any exposed deployment at immediate risk of complete system compromise.

## Quick answers

**Is CVE-2026-105135 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-105135 being exploited in the wild.

**How severe is CVE-2026-105135?** Critical severity, with a CVSS score of 10, published 4 October 2026.

**Which products and versions are affected?** Shanghai AI Laboratory MindSearch 0.1.0 and earlier.

**Is there a fix for CVE-2026-105135?** No fixed version is recorded yet. Check the vendor's advisory in the references below. If you run it, Junglewise drafts a mitigation that fits your setup and applies it once you approve, with senior security engineers behind it.

**How is CVE-2026-105135 attacked?** It is attacked over the network, in a software library.

**Am I affected by CVE-2026-105135?** Check whether you run Shanghai AI Laboratory MindSearch at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

## Technical details

The planner agent in ExecutionAction.run extracts and executes Python code from model responses using unsandboxed exec() with full globals and builtins. An attacker sends a request to the unauthenticated POST /solve endpoint with crafted inputs that cause the planner LLM to emit malicious Python in an interpreter action block. The code is extracted from markdown fences and executed server-side without any sandbox, AST allowlist, or authentication gate, providing full code execution as the server process user (typically root in Docker).

## Affected products

- Shanghai AI Laboratory MindSearch 0.1.0 and earlier

## Timeline

- 2026-08-22: disclosed: Proof of concept disclosed via GitHub Gist
- 2026-10-04: advisory

## References

- [https://gist.github.com/DReazer/7ad46df9da73d0cf414276e4195b2183](https://gist.github.com/DReazer/7ad46df9da73d0cf414276e4195b2183)
- [https://vuldb.com/cve/CVE-2026-105135](https://vuldb.com/cve/CVE-2026-105135)