A new kind of browser puts an AI assistant beside every page you open. It can read the page, sum it up, answer questions about it and, in the more ambitious versions, act on it: fill in a form, click through a checkout, book a meeting.
ChatGPT Atlas and Perplexity's Comet are browsers built around that idea, while the browsers your people already use, Edge and Chrome, are adding their own AI features. By now, some of your people will have tried one.
Why a browser is different
A browser sees more than any other tool on a work laptop. It holds the logged-in sessions for email, your systems, client portals, banking and HR, and an AI browser's assistant can read whatever those pages show. When it acts, it acts with those logins.
It also reads everything, including pages written by people outside your business. Any of them can carry text written to look like an instruction to the assistant, so the prompt injection risk set out in AI security applies on every site a person visits.
What Gartner advised
In December 2025 Gartner published advice titled "Cybersecurity Must Block AI Browsers for Now", which recommended that security teams block all AI browsers for the foreseeable future. Its analysts pointed to defaults that put convenience ahead of security and to sidebars that can send open tabs and browsing history to the vendor's servers.
They also named agents that could make costly mistakes in procurement, travel or HR systems, and the risk of an agent being hijacked to send data to an attacker. For any business that approves one anyway, they advised hardened settings, clear rules for the people using them and continuous monitoring.
Four checks before one is installed
- What does it send? Find out whether the assistant sends page content, open tabs or history to the vendor, and under which terms: a personal account's or a business agreement's.
- What can it do? Check whether it can act inside logged-in sessions, and whether that can be switched off so it only reads.
- Who controls it? A browser your IT team manages through policy is a different proposition from one each person installs and configures for themselves.
- What can it reach? Check whether it can be kept away from the sites that matter most, such as banking, HR systems and client portals.
The AI already in the browsers you manage
Edge and Chrome are adding AI features of their own, and both can be managed through the browser policies your IT team already sets. That is usually the simpler route: decide which features are switched on, for which groups, under which account, and review them as each release changes what is available. Managed this way, AI in the browser stays inside controls you already run.
A sensible position for now
For most businesses of 50 to 500 people, the position that holds up treats the two kinds of AI browser differently.
| Browsers built around an AI agent | AI features in the approved browsers | |
|---|---|---|
| Examples | ChatGPT Atlas, Perplexity's Comet | Edge and Chrome |
| Approved for work devices | No | Yes, after the four checks above |
| How it is enforced | Device management stops them being installed | IT switches them on for the groups that need them |
The position is reviewed at each policy review, because the products are changing quickly and a browser refused now may pass the four checks later.
Where this fits
On devices your business manages, the readiness snapshot can show which AI browsers and extensions are already installed. Setting the browser rules is part of Govern, and watching for new unapproved tools as they appear is part of Trackside.