One of the key themes coming from this year’s Risk Management and Security Summit, held by Gartner, is that the battle against cybercrime and data theft is as much about resilience and detecting breaches as preventing them. Having the right tools in place to detect a breach or some sort of unauthorised activity is critical.
Mike Reagan, the Chief Marketing Officer from LogRhythm, says ransomware is a great example of a threat vector that mandates rapid detection and response and that relies on machine learning and automated response.
“Trying to rely on hunting to detect the presence of ransomware - you’ll be much too late. Ransomware executes in seconds and minutes, not hours, days or years”.
The reality, says Reagan, is that the bad guys will get in. So it’s a matter of accepting that reality and focussing on continuous improvement in speed of detection and rate of response. While there’s a place for “hunting and manual searching for threats but the real value for speed of detection and response is through automation”.
By achieving a high level of automation in response and detection, Reagan says you can then reach a high level of resilience.
“Just because they [threat actors] get in, it doesn’t mean they’re destined to get away with the goods or to cause service disruption,” he says.
This is leading to a shift in how security budgets are being spent. Citing data from Gartner, Reagan says less than 10% of IT security spend in 2013 was on monitoring and detecting technologies. By 2020, that’s expected to rise to 60% - signalling a shift from the “bigger, stronger walls” security strategy to an intelligence-based approach.
The frequency with which these breaches are occurring is driving organisations to accept that reality and adjust their spend and their strategy. But, while prevention is not sufficiently effective today, it’s by no means meant to be ignored. There still needs to be prevention measures. But the distinction to be made between the mindset of three years ago and the mindset of today is you can’t rely so heavily on prevention. There needs to be a blend,” says Reagan.
One of the significant challenges, says Reagan, is “event fatigue”. With so many systems presenting logs, each with its own alerts and notifications, security teams are challenged by trying to find the threat needle in the operational haystack. The trick, he says, is being able to corroborate events from different systems to determine what is going on.
For example, many ransomware attacks start, not by encrypting files, but by seeking and deleting backups. By searching for different types of anomalous behaviour, it becomes possible to detect a threat before it takes hold.
This approach relies on machine learning. However, teaching a system to know the difference between expected and aberrant behaviour can be challenging. Reagan says it’s not just a matter of looking for specific actions but defining parameters, or boundaries, within which “normal” actions occur. When something occurs outside the boundaries the action, in itself might not be an indicator of a threat but when correlated with other out-of-boundary events could signal an attack in progress.
Machine learning can do this far faster than a human can as it’s capable of looking at more data sources and processing them at a greater velocity than humans.
“It takes a lot of compute power. There are PhDs in analytics that understand mathematical techniques for large datasets to pull out those abnormalities. That needs to be part of the equation. There’s also statistical analysis and corroborative analysis”.
Reagan says “It can’t all be machine learning. It needs to be a combination of machine learning coupled with parameter based analytics that will give you the precision of insight most organisations are looking for”.