Norton launches Safe AI protections for chat tools

Norton has launched Safe AI in Norton 360, a set of features designed to monitor how AI chat applications interact with a user's device.

The release adds three protections for consumer AI use on Windows and Mac. The tools are intended to block unauthorised actions, limit access to personal files and flag unsafe links surfaced through AI assistants.

The product arrives as security groups track a rise in threats delivered through generative AI services. Gen Threat Labs said it had recorded a 54% increase in malicious content served to Gen users through AI chatbots, including ChatGPT and Claude, over the past 90 days.

Gen's consumer research also points to unease about autonomous AI tools. According to the company, more than two-thirds of people are concerned about privacy, data breaches or losing control when using an AI assistant that can act on their behalf.

Three tools

Safe AI sits within Norton 360's scam protection and is built around three functions.

The first, Action Shield, is designed to stop unwanted AI actions and block potentially harmful prompt injections before they are carried out. Norton said this reflects growing concern that AI systems are being given broader authority to perform tasks on behalf of users, creating openings for abuse if they are manipulated.

The second feature, Safe Zone, lets users decide which connected files and content an AI tool can access. Permission settings include full access, read-only access or a block. The control is intended to help users manage what personal material AI systems can open or process.

The third function, Safe Links, checks links shared through AI assistants before a user opens them. AI-generated responses can include links that appear trustworthy but lead to harmful destinations.

Support varies by feature. Safe AI is available on Windows and Mac, while Safe Zone is available only on Windows. Safe Links currently supports browser versions of ChatGPT and Claude. Action Shield currently supports Claude, Cursor, OpenClaw and VS Code.

Growing concern

The launch reflects a broader shift in cyber security as AI assistants move from answering questions to carrying out actions on devices and across connected services. That has created a new category of risk in which the issue is not only malicious content, but also what an AI tool may be instructed to do once it has access to files, apps or workflows.

Security researchers have increasingly focused on prompt injection attacks, in which hidden or malicious instructions are inserted into content an AI system processes. In practical terms, that can lead a system to reveal information, follow unauthorised instructions or interact with files and services in ways the user did not expect.

Norton's approach is to place a control layer around those interactions rather than rely solely on the AI provider to detect abuse. That means watching actions on the device, setting boundaries on file access and screening links generated in AI conversations.

Dagim Girma, GM for Norton, framed the issue as basic safety controls struggling to keep pace with consumers' rapid adoption of AI assistants.

"Most of the AI tools people are using today should really come with a warning label that safety isn't included. The tools were designed around how to get the job done and not necessarily how to get the job done right, or even safely, despite the best intentions," said Girma, GM for Norton.

He said the company sees a gap between the speed of AI adoption and the average user's ability to assess new forms of digital risk.

"That's where we come in. People shouldn't have to become AI security experts to benefit from AI tools. With Safe AI, we're helping people stay in control of their information and embrace the possibilities of AI with greater confidence," Girma said.

The launch also underlines how established consumer cyber security brands are adapting products built for malware, scams and identity theft to a market shaped by AI agents and chat interfaces. Rather than treating AI as a separate category, vendors are beginning to fold AI-specific controls into broader security suites used on personal computers and devices.

For Norton, Safe AI extends Norton 360 into one of the fastest-changing parts of consumer software, where users increasingly ask AI systems to search, summarise, open, connect and act.