Beloved browsers like Chrome, Safari, and Firefox may soon join Internet Explorer and the long-forgotten Netscape Navigator in the digital retirement home. A new generation of AI-powered browsers is taking its place, and while many professionals are asking, “How can this optimise my workflow?”, corporate IT officers are asking an even bigger question: “Can we trust this?”
What AI Browsers Do
Let’s answer the first question first. AI-powered web browsers are enhanced to assist, automate and personalise your browsing experience far beyond traditional search and navigation. While Internet Explorer, for example, simply decoded HTML and displayed websites, AI-enhanced Microsoft Edge has Copilot integration which enables it to actively summarise, research, complete tasks and interact with content in real time.
And while Edge is impressive, it’s a relative laggard. Like Opera’s Aria, it’s still a traditional browser with AI sidebars. The cutting edge of AI browsers includes OpenAI’s Atlas (powered by ChatGPT), Perplexity’s Comet, and Dia. It’s unnerving when you use one of these truly AI-powered browsers for the first time.
Take Comet, for example. The browser’s home page looks and feels a lot like Copilot, with a task bar where you can type in your request. There, you can ask Comet to scan your inbox and find emails from the last seven days that require replies and draft responses for each. You’ll then watch as Gmail opens in a tab, and the browser goes through your unopened emails and automatically starts generating responses to each. Granted, the replies are in generic corporate speak, but it all happens without your hands touching the keyboard.
You could also ask Comet – or Atlas or Dia (both of which are currently available for Mac iOS only) – to “Spin up a quick pitch workspace for a slide deck with free images”, and within seconds you’ll have new browser tabs open for Google Slides and the royalty-free image library Unsplash.
Browsing Safely
It’s breathtaking. And it leads to that second question about security. AI’s “agentic” capabilities (when it clicks and types for you) introduce a whole new world of vulnerabilities. You’re giving your computer autonomy, allowing it to perform complex actions on your behalf – often without your direct approval.
Now you – or your company’s IT team – don’t just worry about protecting your laptop from malicious code; you also have to worry about protecting yourself from an AI assistant that could get duped into giving away your private data or your company’s top-secret strategy, while you’re at the watercooler waiting for Atlas to write your TPS Report for Mr Lundbergh.
“Smart” doesn’t automatically mean “safe”. Researchers at the privacy-focused browser Brave recently set up a test asking Comet to summarise a Reddit page. A comment on the page contained malicious commands in hidden text, which led the AI to navigate to a window where the user’s Gmail was open and steal information. The phenomenon is called “indirect prompt injection”, and it’s a security nightmare.
Another nightmare? For an AI browser to act as your assistant, it needs to see and analyse everything you do online (just like any ChatGPT would). How much personal or business information are you comfortable sharing with an outside tech company? And while you could switch off the default training model that shares your information to learn from your behaviour, how much of what’s “private” will really be private – or protected?
Good question. You should ask your browser.
