Norton Adds 'Agent' Protection
Norton expanded Norton 360 with an "AI Agent Protection" feature designed to secure autonomous AI agents running in real time. The announcement signals security vendors are building controls specifically for autonomous workflows as they move from experiments to production. Norton framed the feature as real‑time protection for live AI agents. (prnewswire.com)
A normal chatbot waits for you to click send. An artificial intelligence agent can write code, install tools, move files, and call other services on its own, which turns one bad instruction into a chain of actions on your computer. (support.norton.com) Norton says its new feature watches those live actions in real time inside tools including Claude Code, Cursor, and OpenClaw, and tries to stop unsafe steps before they happen. The company is putting it inside Norton 360 instead of shipping it as a separate lab demo. (support.norton.com) That is a different problem from the scam filtering Norton has been pushing since 2024. Scam protection looks at fake texts, calls, emails, and deepfake videos aimed at a person, while agent protection is aimed at software that can touch your files and run commands. (us.norton.com) (prnewswire.com) Norton’s own support page uses a simple warning: today’s agents can run commands, install software, and interact with your files. That means the risk is no longer just “did the model answer wrong,” but “did the model do the wrong thing on a real machine.” (support.norton.com) The timing tells you where the market is moving. In March 2026, Norton put its Genie scam assistant directly inside ChatGPT, and in April 2026 it is talking about protecting autonomous workflows, which is a step from helping humans judge messages to guarding software that acts for them. (prnewswire.com) (support.norton.com) Security vendors are reacting to the same shift. A recent Arkose Labs study said companies expect agentic artificial intelligence incidents soon, but are still spending only about 6% of security budgets on stopping them. (securityinfowatch.com) The threat picture is already getting more expensive. The Federal Bureau of Investigation’s Internet Crime Complaint Center said reported cybercrime losses reached $20.88 billion in 2025, and recent coverage has tied part of that surge to criminals using artificial intelligence to scale fraud and business email compromise. (forbes.com) What Norton is selling, then, is a seatbelt for software that has started driving. If people are going to let agents book tasks, edit code, and touch local systems in real time, consumer security products now need controls for actions, not just warnings for links. (support.norton.com)
