Facebook announced on Thursday the winners of its 2017 Internet Defense Prize. A team of researchers from the University of California, Berkeley, and the Lawrence Berkeley National Laboratory earned the $100,000 prize for a novel technique of detecting credential spear-phishing attacks in enterprise environments.
The new method, presented these days at the USENIX Security Symposium, combines a new non-parametric anomaly scoring technique for ranking security alerts with features derived from the analysis of spear-phishing emails.
To test their method, the researchers analyzed more than 370 million emails received by a large enterprise’s employees between March 2013 and January 2017.
The first part of the detection method relies on the analysis of two key components: domain reputation features and sender reputation features. The domain reputation feature involves analyzing the link included in an email to see if it poses a risk. A URL is considered risky if it has not been visited by many employees from within an organization, or if it has never been visited until very recently.
The sender reputation feature aims to identify spoofing of the sender’s name in the From header, a previously unseen attacker using a name and email address closely resembling a known or authoritative entity, exploitation of compromised user accounts, and suspicious email content (i.e. messages that reference accounts and credentials, or ones that invoke a sense of urgency).
Once data has been collected about the sender and the URL, the system needs to decide whether or not an alert should be generated for the security team. The method proposed by the researchers, dubbed “Directed Anomaly Scoring (DAS),” ranks all events by determining how suspicious each event is compared to other events. After all the events have been classified, the DAS system selects the highest-ranked events and alerts the security team about their existence.
According to the experts, the new method detected 17 of 19 spear-phishing emails and had a false positive rate of less than 0.005%, which they claim is 200 times lower than what other researchers had previously obtained.
“This research is important for two reasons,” said Facebook’s Nektarios Leontiadis. “First, in recent history, successful spearphishing attacks have led to a number of prominent information leaks. Every time the community improves the detection or prevention of compromise from a technical standpoint, the human factor becomes an even stronger focal point of adversaries. Helping protect people from social engineering attacks becomes even more important. This research can help reduce the potential of such compromises happening in the future.”
“Secondly, the authors acknowledge and account for the cost of false positives in their detection methodology. This is significant because it factors into the overhead cost and response time for incident response teams,” Leontiadis added.
Facebook also announced that two other groups earned honorable mentions for their research on preventing dangling pointer flaws and the use of static analysis techniques to find Linux kernel driver vulnerabilities.
Related: Facebook Awards $100,000 for Post-Quantum TLS Security Research
Related: Facebook Rewards Researchers for Bad Casting Flaw Detection Technique