Nvidia's latest app update, version 11.0.9.565 released on October 13, introduces a feature that allows compatible AI agents to control the Nvidia application. This new functionality, detailed in the update's patch notes, enables AI assistants like Hermes, OpenClaw, or Perplexity to adjust optimal game settings or enable recording features through natural language commands. The feature is facilitated by a local MCP server (Model Context Protocol) integrated into the Nvidia App.
This integration aligns with Nvidia's recent launch of its RTX Spark platform, designed for AI agents on laptops. However, the AI control setting is not exclusive to RTX Spark systems and is present for any user updating to the latest version of the Nvidia App. This has prompted some users to express concerns regarding potential security risks and privacy implications of granting AI control over graphics card functionalities.
Agustin Cordes, cinematic director for Silent Hill: Ascension, voiced these concerns on social media, stating that the setting is enabled by default and represents a "high security risk." Other users echoed these sentiments, with one cybersecurity professional advising the general public to disable the feature, citing it as an unnecessary threat vector. While no exploits have been reported yet, the new connection point could potentially pose future risks.
This is not Nvidia's first foray into AI-assisted GPU management. The company previously revealed Project G-Assist at Computex 2024, an LLM designed to help users manage GPU settings such as clock speeds and fan curves. The primary point of contention for many users appears to be that the AI agent control setting is activated automatically upon updating the Nvidia App.
To disable the feature, users can navigate to the Nvidia App's settings, locate 'AI Agents and App Connections,' and toggle off 'Allow Agents to control Nvidia App.' Users who prefer manual driver installations or wish to avoid potential security vulnerabilities are advised to take this step.