RESOURCES / THE EVOLUTION BLOG
In-browser AI agents: what they are and how to spot them
Alisdair Faulkner
What is an in-browser AI agent?
An in-browser AI agent is an AI assistant that is built into a person's browser or takes control of it. The person asks for a task, such as "find a flight and fill in the booking form", and the agent carries it out in an open tab. The person can watch, step in, or leave the agent to finish.
Because the agent works in the person's own browser, it uses their cookies, saved logins and open sessions. To your site, the session starts as the customer's session. Partway through, software takes over the clicking and typing.
How are in-browser agents different from other agent types?
In-browser agents share more with the real customer than any other type. The device, the network and the browser are all genuine.
A browser extension agent also runs on the real device, but it is a separate add-on that acts from inside the page. An in-browser agent comes from the AI assistant itself. A cloud agent does similar tasks, but from a remote browser on a cloud network, so the session never touches the customer's device.
| IN-BROWSER AGENT | BROWSER EXTENSION AGENT | CLOUD AGENT | AGENT SDK | |
| WHERE THE BROWSER RUNS | The person's computer | The person's computer | The vendor's cloud | A developer's computer or a hosted service |
| NETWORK | Home or office | Home or office | Cloud hosting | Varies |
| WHO CONTROLS THE ACTIONS | The AI assistant the person uses | A separately installed extension | A hosted service | A developer's code |
| HOW A TASK STARTS | The person asks in the browser | The person asks through the extension | The person asks in an app or chat | A script runs |
What are examples of in-browser agents that Darwinium detects?
The two most used in-browser agents Darwinium sees are Claude in Chrome made by Anthropic and ChatGPT in Chrome from OpenAI.
What are the tells of an in-browser agent?
Most device checks see nothing wrong. The device is real, the network is the customer's home or office, and the fonts and settings are normal. The difference shows in how the page is used:
- Clicks jump straight to the target, with no pointer movement before them.
- Form fields are filled in by script instead of typed.
- The agent leaves its own markers in the page.
No single sign is proof. Autofill, for example, also fills fields without typing. Darwinium looks for several signs that agree before it calls a session an agent.
Why does this matter?
In-browser agents usually act for the account owner, using that person's logged-in session. Most of the time the task is legitimate, like a purchase or a booking. You still need to know when an agent, and not the person, is about to pay, change account details or accept terms, so you can apply the policy you want at those steps.
Frequently asked questions
Can an in-browser agent use a customer's logged-in session?
Yes. It works in the person's own browser, so it has the same cookies and logins.
Do device checks catch in-browser agents?
Usually not. The device and network are real, so the signs show up in behavior on the page instead.
Is an in-browser agent the same as a browser extension agent?
No. An extension agent is a separate add-on that acts from inside the page. An in-browser agent comes from the AI assistant itself and drives the browser.