Gen Digital has introduced Safe AI, a new layer of protection built into Norton 360 for Windows and Mac that gives users direct oversight of what artificial intelligence tools are allowed to do on their devices. The feature focuses on three areas that have become pressing concerns as AI assistants gain broader system access: the actions these tools can take, the files they can read or modify, and the links they generate or follow on a user's behalf.

Why AI Tools Have Become a New Attack Surface

AI assistants embedded in browsers, productivity software, and operating systems increasingly operate with permissions that once belonged exclusively to the human user. They can open files, summarize documents, draft messages, and click through web pages without a person directly steering each step. That autonomy is useful, but it also creates a fresh vector for exploitation. A malicious actor who manipulates an AI tool, through a poisoned document or a deceptive prompt embedded in a webpage, can potentially trick the system into exposing sensitive data or navigating to a harmful destination without the user ever realizing the tool was compromised.

This risk category, sometimes discussed under the umbrella of prompt injection or agentic misuse, has worried security researchers for the past two years as AI agents moved from experimental labs into everyday consumer software. Traditional antivirus tools were built to catch malicious code, not to judge whether an AI assistant's behavior falls outside expected boundaries. Safe AI is Gen's attempt to close that gap by treating AI activity as its own distinct category requiring monitoring, much like network traffic or file downloads have long been monitored.

What the Protections Actually Control

According to Gen's announcement, Safe AI gives Norton 360 users the ability to set boundaries on three fronts:

• Actions: limiting what automated tasks an AI tool can perform on a device without explicit confirmation.
• File access: restricting which documents and folders AI applications are permitted to read or alter.
• Links: flagging or blocking URLs that AI tools generate or attempt to open, reducing exposure to phishing or malicious redirects introduced through automated browsing.

The approach mirrors the logic of long-standing endpoint security principles, least privilege and explicit consent, applied specifically to the new behavior patterns of generative and agentic AI software rather than conventional applications.

An Early Move in a Fast-Shifting Market

Gen's decision to fold these controls directly into Norton 360, rather than selling a separate product, suggests the company sees AI-related risk as a baseline expectation for consumer security suites going forward, not a premium add-on. That positioning makes sense given how quickly AI features have been built into mainstream software that ordinary consumers already use daily, often without users fully understanding what permissions they have granted.

What remains unclear is how this expansion will affect Gen's business in practical terms. The announcement does not disclose adoption figures, whether Safe AI will shift pricing across Norton 360 tiers, or whether it is expected to improve subscriber retention. Security vendors have historically used new feature categories, firewall improvements, VPN bundling, identity theft monitoring, to differentiate subscription tiers and justify renewals. Whether Safe AI follows that pattern, becoming a selling point for higher-priced plans, or whether Gen treats it as a standard feature across its consumer base, will likely become clearer in future earnings disclosures rather than in the initial rollout.

What Users Should Understand Going In

Consumers should view Safe AI as a control layer, not a guarantee. No security feature can fully anticipate every method by which an AI tool might be manipulated, particularly as attackers adapt their techniques to new defenses. The value of granular permission settings lies in reducing the blast radius of a compromise, limiting what a manipulated AI assistant can actually reach or do, rather than promising to prevent manipulation outright. For now, Safe AI represents an acknowledgment from one of the industry's largest consumer security vendors that AI tools require the same scrutiny once reserved for suspicious downloads and unfamiliar networks.