AI agents: When bots browse the web

Digital generated image of robot's hand holding credit card against blue background — 'It makes sense for AI companies to jump into the browser game' (Image credit: Andriy Onufriyenko / Getty Images)

The battle over the future of web browsing is here, said Shirin Ghaffary and Matt Day in Bloomberg. Amazon last week sued the artificial intelligence startup Perplexity because its new AIpowered web browser, Comet, can “make purchases on a real person’s behalf.” The world’s largest online retailer says this amounts to “computer fraud” when not disclosed. The clash between the two companies offers “an early glimpse into a looming debate” over “agentic artificial intelligence.” Perplexity is among several tech firms, including Google and OpenAI, racing “to rethink the traditional web browser around AI,” with automated agents that can complete tasks like emailing or shopping. Amazon, which is developing its own AI-powered shopping agents, has reason to worry: If more bots do the shopping for humans, that poses “a significant threat to Amazon’s lucrative advertising business.”

It makes sense for AI companies to jump into the browser game, said David Pierce in The Verge. Your browser holds “a vast trove of data about you”—including everywhere you go online, and what you do there—which can used to precisely target ads that generate revenue. And it also “contains the most important input system on the internet,” a box to do Google searches. “If AI interactions are going to usurp Google searches, they have to be that easy.”After testing several AI browsers, I’m a convert, said Nicole Nguyen in The Wall Street Journal. The best part of such a browser is that it has “a built-in chatbot that can see what’s open in your tabs.” You can type questions, like “Is this the best price?” and it will “instantly understand the context” and complete tasks based on the answers. I’ve even let OpenAI’s new browser, Atlas, shop for cheap flights on its own “while I did other stuff.”

It’s risky “letting AI this deep into your life,” said Geoffrey A. Fowler in The Washington Post. AI agents “are still prone to mistakes—and when an agent has access to a browser with your login credentials and payment info, that’s a lot of power to hand over.” It also “brings privacy risks that are hard to understand, much less control.” OpenAI’s Atlas “doesn’t just log which websites you visit; it also stores ‘memories’ of what you look at and do on those sites,” going a step beyond traditional cookies. Such agentic systems are ripe for abuse by cybercriminals, said Hiawatha Bray in The Boston Globe. All a hacker needs to do is “hide malicious code inside a webpage” that a bot might pull up. If the code tells my browser to open my password management system, thieves could have “total access to my banking and credit accounts.” For now, “sticking to my dumb old browser seems like the smart move.