We have heard a lot this year on AI allowing new scams, from to. However, a new report suggests that AI also presents a risk of fraud in the other direction – easily falling for scams that human users are much more likely to catch.
The report, comes from a cybersecurity startup called Guardio, which produces a browser extension designed to catch scams in real time. Its conclusions concern so -called “AI AI” browsers as, who travel internet for you and come back with results. The ADIA AI claims to be able to work on complex tasks, such as the creation of a website or the planning of a trip, while users turn.
There is a huge problem here from the point of view of security: although humans are not always excellent in sorting the fraud of reality, AI is even worse. An apparently simple task like summarizing your emails or buying something online comes with a myriad of slippage possibilities. Lack of common sense, agentic AI can be subject to stifling obvious traps.
Guardio researchers tested this hypothesis currently using the only widely available agent browser. Using a different AI, they turned a false website pretending to be Walmart, and then have access to Comet to buy an Apple watch. Ignoring several clues according to which the site was not legitimate, including a manifestly wobbly logo and URL, Comet finished the purchase, putting financial details in the process.
In another test, the authors of the study sent themselves by e-mail claiming to be Wells Fargo, containing a real phishing URL. Comet opened the link without lifting alarms and cheerfully vile a username and a bank password on the Phishing site. A third test proved to be comet likely of an injectable scam, in which a text box hidden in a phishing page ordered AI to download a file.
It is only a set of tests, but the implications give to think. Not only are agentic AI browsers likely new types of scams, but they can also only be vulnerable to the oldest scams in the book. The AI is designed to do what its prompter wants, so if a human user does not notice the signs of a scam the first time he looks, the AI will not serve as a railings.
This warning occurs while each leader in the field is betting very well on agentic AI. Microsoft is, Openai, and Google has been underway since last year. If the developers do not start to integrate better detection of the scam into their navigators, the risks of agental AI become at best a massive blindness – and a new vector of attack at worst.
