Nvidia launches security platform for AI agents

What happened

Nvidia (a company that makes the computer chips most AI tools run on) announced a new security platform on 28 September 2026 that the company says can stop AI agents (software that can take actions on its own, not just answer questions) from going rogue. According to the Guardian, Nvidia said the system was designed to prevent incidents like those that have happened at top companies. The same day, Nvidia also announced a 150 billion dollar stock buyback, the largest in US corporate history.

The article does not explain how the platform works, what incidents it refers to, or when it will be available.

What it means for a small business

AI agents are already being used to book meetings, summarise documents, write code and reply to customer messages. The difference between an AI agent and the chatbot you might use on ChatGPT is that an agent can actually do things: send an email, update a spreadsheet, post to social media, place an order. That makes them useful, but it also means the consequences of a mistake are real.

Say you run a small online shop and you set up an AI agent to reply to customer questions about delivery times. If the agent misunderstands a question and promises next-day delivery when you cannot provide it, you have a customer service problem and possibly a refund to process. If an agent with access to your accounts system makes an error, the stakes are higher.

Nvidia's announcement is a sign that the companies building AI infrastructure know these risks exist and are working on ways to manage them. But a security platform from a chip company will not fix every problem. The question for a small business is still the same: what is this tool allowed to do, and what happens if it does the wrong thing?

Three things to try this month

Set clear boundaries before you automate. If you are using or planning to use an AI tool that can take actions, write down what it is allowed to do and what it is not. For example, can it send emails on your behalf, or only draft them for you to review?

Test in a safe space first. Before you let an AI agent loose on real customer data or live systems, run it in a test environment. Send it sample questions, give it fake data, and see what it does. If you do not have a test environment, start with tasks that are easy to undo.

Check what happens when it goes wrong. Work out how you will know if the agent makes a mistake, and what you will do about it. Do you get an alert? Can you reverse the action? Who is responsible for fixing it? You need to make those business decisions before something breaks.

If you are wondering where else AI or automation could help your business, take the free What could your business automate? check.