Do I have to be careful with AI agents?

What exactly is an AI agent?

A web browser equipped with artificial intelligence (AI). “In contrast to traditional browsers, which search the web for information and then merely list it, agent-based browsers go one step further,” explains Marcel Waldvogel, an IT security expert at Migros. “They act like personal assistants, working independently on behalf of the user.”

What specific tasks can AI agents perform?

AI agents can make purchases or compare different hotels and then book skiing holidays independently. They can also make appointments or retrieve and write e-mail. However, to do so, users must first disclose their credit-card details and provide access to their e-mail inbox, calendar and preferred online shops. But caution is advised!

Where can I find these browsers and how much do they cost?

All major AI providers have or are developing their own browsers. For example, OpenAI has launched ChatGPT Atlas, while Perplexity has the Comet browser. Google uses Gemini whereas Microsoft is constantly working on new versions of Edge. There are some completely new providers too, like Fellou. To use this technology, users must download the browser or install an extension. A subscription costing about CHF 20 a month is often required.

What are the risks?

It’s well known that AI can hallucinate and provide information that sounds plausible, but is in fact wrong. The AI agent may then use such false information to make unwanted transactions. What's more, websites can intentionally be designed to fool AI browsers. As a result, the browser could download malware or publish inappropriate posts without the user's knowledge - but for which they can be held responsible. "There’s also a risk of manipulation by browser manufacturers, who need revenue streams to finance their activities. Product placement in browsers will soon be possible. AI browsers could then give preference to the brand that booked the most advertising space instead of a cheaper or higher-quality provider,” Marcel Waldvogel warns.

Can the risks be minimised?

“At the moment, I’d recommend sticking to AI-powered search functions like ChatGPT, and not switching to AI agents just yet. This allows users to retain control over their activities,” the Migros IT security expert advises.