Executive Summary
The Hacker News reported an NVIDIA NemoClaw weakness letting a malicious web page take unauthenticated control of the local Ollama instance serving an AI agent, and a Marimo notebook code-injection flaw (CVE-2026-75149) that allowed attacker-supplied MCP commands in crafted notebooks.
What Happened
The Hacker News reported a vulnerability in NVIDIA NemoClaw that could allow a malicious webpage to take unauthenticated control of the local Ollama instance serving an AI agent and insert hidden instructions into the model. It also reported a Marimo notebook code-injection flaw that allowed attacker-supplied MCP commands in specially crafted notebooks, tracked as CVE-2026-75149.
Affected Products
- NVIDIA NemoClaw
- Ollama
- Marimo
Vulnerability / Attack Technique
Unauthenticated local model control; MCP command injection via crafted notebooks
Technical Mechanism
A web page could reach the local Ollama server without authentication and insert hidden instructions into the model. Separately, specially crafted Marimo notebooks could carry attacker-supplied MCP commands that were executed.
AI Security Category
LLM03:2025 Supply Chain
Part of the MCP Security topic.
Part of the AI Supply Chain and AI SBOM topic.
Relevant Frameworks
RealGround Analysis
Both issues sit in the tooling around agents rather than the model. RealGround analysis: local model servers and notebooks that speak MCP are part of the agent's attack surface and need authentication and input validation.
Recommended Controls
- Require authentication on local model servers and block cross-origin access
- Update Marimo to a fixed version
- Treat notebooks from untrusted sources as code and review MCP commands before execution
Related RealGround Service
AI Agent Business Logic Audit · or start with the free AI Agent Security Scanner.
Sources
The Hacker News: https://thehackernews.com/search/label/AI%20Security?max-results=20&m=1&hl=ko_KR
Summary and analysis written by RealGround from the source above; last updated 2026-10-17.