
🤖🚨 CRITICAL CODE-INJECTION FLAW DISCLOSED IN OPEN-SOURCE AI AGENT FRAMEWORK A new critical vulnerability has been disclosed in: InternLM MindSearch CVE-2026-105135 CVSS v3.1: 10.0 / CRITICAL MindSearch is an open-source AI search/agent framework that uses multiple agents to plan searches, execute actions and synthesize answers. The vulnerability affects: MindSearch 0.1.0 and sits inside the: Planner Agent ↓ mindsearch/agent/graph.py ↓ http://ExecutionAction.run() According to the vulnerability disclosure, attacker-controlled manipulation of the: inputs argument can result in: CODE INJECTION ↓ ARBITRARY CODE EXECUTION The attack can reportedly be initiated REMOTELY. Security properties reported for the flaw: * Network reachable * Low attack complexity * No privileges required * No user interaction required * High confidentiality impact * High integrity impact * High availability impact An exploit has reportedly been disclosed alongside the vulnerability. However: * No confirmed exploitation in the wild has been identified * CVE-2026-105135 is NOT currently in CISA KEV * No fixed MindSearch release has been identified in the current CVE record * Vendor response/remediation remains unclear ⚠️ Analyst Note: This is exactly the security boundary AI-agent platforms need to get right. Agent frameworks increasingly take: UNTRUSTED INPUT ↓ LLM REASONING ↓ PLANNER ↓ EXECUTION ACTIONS ↓ HOST ENVIRONMENT If attacker-controlled content crosses from the reasoning layer into executable code without a hard security boundary, the AI agent itself can become the path to host compromise. This vulnerability was publicly disclosed: OCTOBER 4, 2026. Organizations experimenting with exposed MindSearch deployments should restrict access until the vendor provides definitive remediation guidance. Upstream vulnerability disclosure: https://vuldb.com/vuln/413352 InternLM MindSearch: https://github.com/InternLM/MindSearch #MindSearch #InternLM #AISecurity #AgenticAI #CVE2026105135 #RCE #CyberSecurity #DDW
Post summary
The text announces a critical code injection vulnerability in InternLM MindSearch (CVE-2026-105135) with a CVSS score of 10.0, providing technical details and noting that an exploit has been disclosed but no active exploitation or patch is available.









