Nvidia's software update supports Agent Control, but some users are warned to turn it off

It's no secret that Nvidia loves its AI, and fresh from the launch of RTX Spark – the new chipset for Agentic laptops – Nvidia's app now has a trigger that lets AI agents take control of it. The problem is that it is on by default.

It's no secret that Nvidia loves its AI, and fresh from the launch of RTX Spark – the new chipset for Agentic laptops – Nvidia's app now has a trigger that lets AI agents take control of it. The problem is that it is on by default.

Nvidia added this option on October 13 in Nvidia App 11.0.9.565. In the patch notes for that update, it says: "The Nvidia application includes a local MCP (Model Texture Protocol) server that controls the Nvidia application using natural language requests to a compatible AI agent. settings, or enable Shadowplay to record gameplay, all through simple conversation."

It's no surprise that Nvidia has added this option to the Nvidia app. After all, it just announced its RTX Spark platform, which is built entirely around AI agents, so it makes sense that Nvidia's app would allow users to import those agents. But this setting is not exclusive to RTX Spark systems and will be shown to anyone who installs the latest version of the app.

And of course, some people worry about the security implications of letting an AI agent control their graphics card.

Silent Hill: Ascension Cinema director Agustin Cordes tweeted about the update: “PSA: Can confirm that the Nvidia app updates automatically and *by default* has enabled settings that allow AI agents to control the app with very lenient rights. "In addition to harmful privacy. Invasion features and poor AI is a high security risk that you may not want."

Meanwhile, a number of fans quoted his post, such as user TheAcidBurns3 saying: "Please pay attention to this. You don't want access to your graphics card like this!"

User DATGUYCOOP, who appears to be a cyber security pro, asked most of the weirdos to turn off the feature, saying, “I'm just saying turn this feature off. Trust from the cyber security man. If you're working with AI agents, it's a good way to add filtering, protection, and secondary checks, but to normal people, it adds a threat vector that's best closed. Currently, there are no unreported exploits yet, but it's new and adds a possible interface that might do something later, so just turn it off."

The problem really seems to be that Nvidia's software update turns this setting on by default, which is pretty much a surefire way to make people hate any particular feature in a tech product. Nvidia is no stranger to letting AI systems and agents screw up your graphics card. Anyway, the company announced Project G-Assist at Computex 2024, which was basically an LLM that helps you manage GPU settings like clock speeds, fan curves, and more.

I've reached out to Nvidia about the security implications of this new Nvidia software update and will update this article if I hear back. But regardless of what Tim Green says, that option is unlikely to go anywhere anytime soon. So, if you don't want your GPU driver to be open to AI agents, either install your graphics driver manually or turn it off by going to the Nvidia app, click Settings, then go to AI Agents and App Connections and disable Allow Agents to Control Nvidia App.