A new study of anti-phishing toolbars has come to a stark conclusion about their effectiveness - none of them are any good.
A week ago, a report from SmartWare told the world that Mozilla’s Firefox 2.0 had the best anti-phishing capabilities, while a month ago a report from 3sharp claimed Microsoft’s Internet Explorer 7.0 was tops.
But however apparently independent their methodology, those reports were sponsored by the companies - Mozilla and Microsoft - that triumphed in each test, a fact that has undermined their reliability in many people’s eyes.
By contrast, the new study [pdf], Finding Phish: An Evaluation of Anti-Phishing Toolbars, was conducted by researchers at Carnegie Mellon University in Pittsburgh, backed by organisations as worthily anodyne as the US National Science Foundation and the US Army Research Office.
The study looked at ten browser toolbars: Microsoft Explorer 7, eBay, Google, Netcraft (Mozilla), Netscape, Cloudmark (Mozilla), Earthlink, Geotrust’s TrustWatch, and Stanford University’s Spoofguard, and McAfee’s SiteAdvisor.
Even the best of the bunch - Earthlink, Netcraft, Google, Coudmark, and Explorer 7 - detected only 85 percent of fraudulent websites, a good but far from secure level of effectiveness. The rest scored under the 50 percent mark, with McAfee’s SiteAdvisor unable to spot any.
"Overall, we found that the anti-phishing toolbars that were examined in this study left a lot to be desired," the authors concluded. "Many of the toolbars we tested were vulnerable to some simple exploits as well."
Most of the toolbars suffered to varying degrees from false positives, where legitimate sites were erroneously identified as being phishing sites. The researchers reckoned this to be almost as big a problem as missing a real phishing site because constant warnings about sites known to be OK might persuade users to ignore all warnings, even when correct.
The inclusion of SiteAdvisor will likely attract the ire of McAfee, which was angered by the product’s dreadful showing in a Microsoft-sponsored test of anti-phishing toolbars carried out some weeks ago by consultancy 3Sharp. McAfee claimed after the event that the product was not intended to perform as an anti-phishing filter, something still disputed by 3Sharp.
Despite the academic thoroughness of their experiments the researchers offered no explanation as to why McAfee’s SiteAdvisor showed zero ability to spot anti-phishing websites, an odd statistic even within the context of the other product’s modest abilities. The fact that the software might not have had any anti-phishing features - regardless of how it had been described on its website - doesn’t appear to have dawned on the team.
The researchers come up with a number of general observations about browser phishing security, the first and most obvious of which is to use all such filters with great care - no toolbar can spot all fraudulent sites to a high enough degree or reliability to make it a firm line of defence.
The second point, referred to only obliquely, is that origin of the browser on which the filter is running - the Mozilla versus Internet Explorer debate - doesn’t appear to make any odds. It is the effectiveness of the heuristics used to identify sites that matters, and the usability of software design from the point of view of the users themselves.