‘It’s a Pivotal Moment’: Why Meta’s New AI Agent Could Make the Internet Messier
Meta’s new AI assistant, Muse, is designed to act on your behalf out in the world, cancelling subscriptions, haggling with customer service, ordering groceries, and soon making phone calls, all without supervision. It’s the first free, fully-featured AI agent from a major tech company, and it has already been downloaded by millions, with OpenAI close behind with an agent of its own. According to the experts interviewed for this piece, this marks the start of a genuinely new, and genuinely messier, chapter for the internet.
The web was designed around human limitations, limited time, limited patience. AI agents have neither. Ramesh Raskar, an associate professor at MIT who studies these systems, compares the moment to releasing driverless cars with no traffic rules: “Imagine there are no rules of the road. And you release billions of cars – the AI agents – and you just let them drive through playgrounds, hitting kids. That’s where we are. It’s a pivotal moment.”
Setting up Muse requires handing over your name, then full access to your Gmail, calendar, and, on a Mac, your entire computer. Patrick Wardle, co-founder of the security nonprofit Objective-See, which uncovered flaws in the app, didn’t mince words: “I wouldn’t use it. Personally, I would tell you to uninstall it altogether.” A Meta spokesperson counters that Muse was “designed to be safe, secure and private – with built-in protections and user controls that put people in charge,” and underwent extensive testing.
A personal scalper bot for everyone
Experts point to ticket scalping as a preview of what’s coming. In 2024, fans flew to Europe to see Taylor Swift because US resale prices had been driven so high. Now, says Calli Schroeder of the Electronic Privacy Information Center, that dynamic could spread everywhere, as personal scalper bots compete for restaurant tables, driving-test slots and appointment bookings. “There’s going to be a huge escalation of the problems we’re already having with bots,” she warns, adding that little planning has gone into managing the fallout: “We’re just launching these tools and saying, ‘well, we’ll deal with the problems when they come up’.”
There’s also a quieter toll on the businesses underpinning the web. AI chatbots already answer many questions directly, routing traffic, and ad revenue, away from the sites that used to provide it; one analysis found bot traffic jumped 124% in the year to June 2026, overwhelming infrastructure never built to handle it. Meta says Muse is built to respect the interests of the websites it interacts with.
Then there’s the question of trust. If an agent books your holiday or recommends music, Schroeder notes there’s no way to verify it chose the best deal rather than one that benefits its maker’s business partners, and she says Muse’s terms of service offer no guarantee it will act purely in the user’s interest. Meta maintains Muse behaves like “a personal assistant acting for a single person” and follows ethical guidelines that put users first.
Raskar believes the solution lies in treating AI models like engines and the agents built on them like cars that need traffic laws, “windshields, brakes, traffic lights and so on.” Some platforms have already begun banning accounts for AI misuse, and the UK recently overhauled its driving-test booking system partly to curb bots. Still, Raskar remains hopeful about what agents could unlock in terms of innovation, provided the rules are fair to everyone. In the meantime, after a week letting an AI named “Bob” order dental floss and negotiate a discount on his behalf, one thing was clear: handing over an email account that holds two decades of medical records, contracts and old love letters is not a decision to take lightly, bank account access was judged a step too far.
Source: BBC News