AI in the browser: Google’s new auto-browsing +++ Firefox with AI “kill switch” | News

Whether it makes sense and is requested by users or not, many manufacturers are currently implementing a wealth of AI functions in their products. The latest extension from Google, for example, is “Auto Browse” in Chrome. This is an AI agent that can take on automated tasks. The bot operates websites, opens tabs, clicks, fills out forms, compares prices – right up to tasks like “please book me a trip on my days off that costs a maximum of 500 euros” or “Look for a new apartment: 2 rooms, under 1000 euros, max. 30 minutes commute time to work”. However, the final step, i.e. completing the purchase or establishing contact, is still the responsibility of the user. Commissioning an AI to spend money and trusting it completely will probably sound unconvincing to most people.

The functions described are initially only available in the USA and only for subscribers to Google AI Pro or AI Ultra. Auto Browse uses Gemini 3 as a model and transfers the content of the page to Google’s servers for processing. This also causes concern, as it means that a lot of personal information comes to Google that you would otherwise not want to give out.

Firefox: Less AI if the user wants it
Mozilla, on the other hand, caters to those users who don’t want to use AI support at all. From Firefox 148 (planned for February 24, 2026), there will be a new “AI Settings” area in the settings. The strictest level “Blocked” means that no current or future AI functions can be seen – local models that have already been downloaded will then even be deleted again.

This positions Firefox quite clearly compared to competing browsers in which AI features are active by default and are often difficult or impossible to switch off completely. This is a reaction to clear user rejection of “AI everywhere” – and the said kill switch is therefore the logical solution, as the official says. Mozilla explicitly emphasizes that everything should be optional and, where possible, they prefer local models instead of cloud AI.