Sign In

Communications of the ACM

ACM TechNews

Spy Research Agency Is Building Psychic Machines to Predict Hacks


View as: Print Mobile App Share: Send by email Share on reddit Share on StumbleUpon Share on Hacker News Share on Tweeter Share on Facebook
A new contest may hold the key to a system for predicting cyberthreats.

The U.S. Intelligence Advanced Research Projects Agency is developing a contest aimed at creating a system that could predict the cyberthreats a network may face.

Credit: Shutterstock

The U.S. Intelligence Advanced Research Projects Activity (IARPA) is working on a new contest that will challenge government and private-sector entities to create a system for analyzing numerous streams of data from social media to black market malware storefronts to create predictions of what cyberthreats a given network may face ahead of time.

The Cyber-Attack Automated Unconventional Sensor Environment (CAUSE) project is envisioned as a cybersecurity equivalent of systems that have been able to analyze various data streams to successfully predict political uprisings and the spread of diseases such as Ebola.

IARPA's Rob Rahmer, who is leading the CAUSE project, says the competition is meant to help move the cybersecurity field from a reactive to a proactive posture. Such a system would not be perfect and would make mistakes, but Rahmer says it would help agencies and businesses spend their cybersecurity resources proactively.

CAUSE is envisioned as a four-year race and IARPA currently is developing guidelines and determining what the prize for the competition will be. There already is strong interest in the project; about 150 would-be participants from the private sector and academia attended a recent informational workshop about CAUSE.

One issue that needs addressing is what computing resources competitors will need to use; CAUSE would likely require supercomputer-level computing power.

From NextGov.com
View Full Article

 

Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account