An anti-phishing startup is rethinking its approach to protecting consumers and businesses from malicious websites with computer vision technology. Pixm's browser plug-in uses artificial intelligence to analyze websites and determine if they're impersonating a legitimate company.
Pixm was founded in 2015 to protect users from browser-based phishing attacks that appear in emails, chats, and social media. Co-founder Chris Cleveland was a graduate student studying machine learning and computer vision, which describes software that can see the same way humans can. He sought to explore how the concept could be applied to block phishing attacks.
At the time, he said, many organizations' anti-phishing tools looked for malicious URLs based on IP reputation or whether or not they'd been involved with a previous attack. Older applications could scan for signatures indicating a website was spoofing Bank of America or Wells Fargo, but only if an attack had already taken place. There was little to detect brand-new phishing threats.
"If you're protecting people by showing them the yellow tape where the old crime scene occurred, you're not actually protecting them from the initial crime," Cleveland says. "The bar is low to launch a brand-new attack. … You can do it very fast, and it takes very little knowledge."
Pixm won a pitch competition and took its tool to The Cybersecurity Factory, a program for security startups run in partnership with Highland Capital. The company was connected with a larger organization conducting anti-phishing in a Web-scanning context, scouring the Web to find attacker infrastructure and identify phishing campaigns. Soon, Pixm launched its initial product, an API that accepted URLs and scanned them for "threat needles in the haystack."
When the trial with the larger organization ended, Pixm decided to take its prototype and build it out as a real-time device application. "We figured that this could really solve the problem on the end-user device, which is really where the threat exists," he adds.
Now, its computer vision technology has been developed into a desktop application with versions for both consumer and enterprise users to detect and alert to browser-based phish. It's platform-independent and available on Chrome, Chromium-based browsers, and Firefox.
Pixm's goal is to approach phishing from an end-user perspective and detect the techniques attackers use to gain trust. Users may see a logo and text that appear trusted but are from a different origin, domain, or IP address. The software is developed based on these psychological cues of trust so it can recognize visual branding associated with commonly phished brands, such as Gmail or Outlook. It currently supports more than 100 brands, Cleveland notes.
"The cool thing about using deep learning and convolutional network tools is that you could perform the same task in real time and you could use actual training data — in our case, from millions of different screenshots and different kinds of phishing attacks — and train algorithms to learn an arbitrary number of patterns and have this run in real time," he explains.
The consumer and business versions of Pixm are built off the same core installable. It's a single-click installation for consumers, Cleveland says. The enterprise version will let admins ensure Pixm is installed across devices in an environment. It will also allow integration with personal devices, given the pattern of attackers targeting victims on both personal and corporate machines.
"For the enterprises as well … in addition to being able to enforce that on the endpoint, you're also going to have centralized visibility," he continues. "So, if Joe from this department clicks on that threat link, you can see where that activity is and you can respond accordingly."
Cleveland acknowledges that companies in different industries have different privacy and infrastructure concerns and that Pixm is building out capabilities to address these. It's also planning to develop a mobile version of the application to address phishing attacks on phones.
Over time, as Pixm expands its user base, it hopes to be a source of threat intelligence for phishing attacks and use this data to continuously improve its product.