14 DAY TRIAL // Scientists trained an AI system to be able to recognize malware hidden in links shared on Twitter but obfuscated by the social network's default URL shortener (t.co).
According to recently published findings, researchers experimented with a complex URL and PC analysis system during this year's Super Bowl and Cricket World Cup final.
This system monitored how a website's content forced changes on a user's machine, changes that are not typically associated with normal Web browsing behavior.
This included modifications to registry files, the creation of new operating system processes, and local file modifications.
The AI looks for early signs of infection
By analyzing how bytes and packets were exchanged between websites and the machine, scientists were able to train the AI system to recognize predictive signals and detect malicious URLs.
With a large number of links being shared during live sporting events and natural disasters, links that normally get shortened by Twitter t.co service, scientists hope to provide Twitter and its users with a system that can protect them from being infected with malware while accessing one of the obfuscated tweets.
Because cyber-criminals will provide insightful and interesting tweets along with their malware-infested t.co URLs, this AI could benefit Twitter by allowing it to weed out misbehaving and abusive bot accounts.
More tests are planned for next year's EURO 2016 championship
For their initial tests, researchers claim they identified potential cyber-attacks after users clicked on a URL posted on Twitter within 5 seconds with an accuracy of up to 83%, and within 30 seconds with an accuracy of 98%.
Further tests are planned for next year's European Football Championship, which will be held in France during the months of June and July.
The study was carried out by research personnel from the Cardiff University and was funded by the Engineering and Physical Sciences Research Council (EPSRC) and the Economic and Social Research Council (ESRC).
This research was presented at the 2015 IEEE / ACM International Conference on Advances in Social Networks Analysis and Mining in August 2015.