An analysis by security researchers of 48,000 extensions for Google's Chrome browser uncovered many that are used for fraud and data theft, actions that are mostly undetectable to regular users.
The study, due to be presented Thursday at the Usenix Security Symposium in San Diego, forecasts growing security problems around extensions as cybercriminals tap into the rich data contained in Web browsers for profit.
They found 130 outright malicious extensions and 4,712 suspicious ones, engaged in a variety of affiliate fraud, credential theft, advertising fraud and social network abuse.
"By installing an extension, you will not see any malicious behavior," said Alexandros Kapravelos, a doctoral candidate at the University of California at Santa Barbara, in a phone interview. "You need to visit specific pages to trigger the malicious behavior."
The researchers developed a system called Hulk, which closely watches how browser extensions behave when interacting with websites. Part of that involved creating so-called "HoneyPages," which are specially crafted Web pages designed to tease out bad behavior.
Because extensions add extra functionality to the browser, they need a lot of power. Extensions often ask for a variety of permissions that come from Chrome's APIs (application programming interfaces).
"That is just way too powerful," Kapravelos said. "We need to be way more careful into which extensions we give these permissions."
During the study, the researchers worked closely with Google. The company does review extensions before they're allowed in Chrome's Web Store, but it hasn't stopped some of the bad ones from getting in.
In light of the study, Google has implemented several changes to strengthen its control over extensions. It is now harder to install extensions from outside the Web Store, a practice known as "side loading," said Chris Grier, a security researcher with the University of California at Berkeley and a co-author of the study.
Very few extensions were found that try to interfere with online banking sessions or logged keystrokes, Grier said. But further analysis might reveal bad behaviors that are well-hidden, he added.
Some of the suspicious extensions they studied have been downloaded millions of times.
One extension aimed at Chinese users that has been downloaded 5.5 million times uses a tracking beacon to report all the Web pages a person has visited to a remote server. Those reports are not sent encrypted over SSL (Secure Sockets Layer).
"While this isn't malicious per se, it certainly exposes Chinese users to a whole new set of risks," Grier said. "Now all of their content is no longer encrypted, and there are no confidentiality guarantees. Even for users outside of China, there's certainly a lot of risk in having every http request reported back to some server."
In another example, some extensions changed or added parameters within a URL in order to accomplish affiliate fraud.
Companies such as Amazon will pay a small fee to webmasters, known as affiliates, when someone clicks a link on their website that leads to a sale. That is tracked by adding an affiliate code inside a URL.
Some extensions will swap out the legitimate affiliate code for their own, effectively gaining fraudulent credit for the sale. Since Google saw the study, it has moved to crack down on affiliate fraud in its policy governing extensions, Grier said.
They also found examples of extensions swapping out advertisements on a website for their own in order to get the fees. Sometimes extensions swapped out banner ads, injected ads into ad-free sites such as Wikipedia or overlaid ads on top of website content.
The study was also co-authored by Neha Chachra, Christopher Kruegel, Giovanni Vigna and Vern Paxson.
Send news tips and comments to firstname.lastname@example.org. Follow me on Twitter: @jeremy_kirk