Intelligent Cyber Defense
Hailed by some as a Holy Grail for cyber security protection, machine learning programs are helping businesses identify and counter cyberattacks more effectively than ever before.
From programs scanning the dark web for clues of cyberattacks to software analyzing companies’ data network flows and user behavior, risk managers, CTOs and CIOs have a growing choice of tools at their disposal.
As more devices come online and more data is produced, the potential vulnerabilities hackers can exploit grow exponentially. So too grows the need for tools that help firms spot threats and strengthen their cyber networks.
“It is increasingly important to develop tools to sift through the noise, identify signals and check for anomalies to identify attack vectors that are susceptible and ‘being exploited,” said Eric Cernak, vice president, Hartford Steam Boiler. “Machine learning can really help this process.”
Leveraging powerful algorithms, programs that harness machine learning are getting better at spotting the difference between genuine threats and innocent anomalies. They can detect threats faster than before.
Ryan Griffin, senior vice president, JLT Specialty USA, pointed out that the average time to detect an event was 180 days. Now it can be done in three.
According to ethical hacker and cyber security expert Mike Peters, vice president for IT, RIMS, more than 70 percent of attacks exploited known vulnerabilities in available patches last year. Machine learning programs can process vast quantities of data while iteratively learning, promising to help uncover many of these threats with minimal human involvement, he said.
“Machine learning systems have a big role to play in conquering threats currently handled in a manual fashion.”
“Machine learning systems have a big role to play in conquering threats currently handled in a manual fashion.” – Ryan Griffin, senior vice president, JLT Specialty USA
Credit rating firm FICO launched its own cyber vulnerability assessment tool, which scans the entire web on a weekly basis, gathering data company internet footprints. It learns about the conditional attributes and vulnerabilities exhibited by companies in the lead-up to a breach. Companies using this software scan and compare their own internet footprints to assess their cyber risk level.
According to Graeme Newman, cyber leader, Barbican Insurance, the program has so far found the most at-risk company to be 24 times more likely to suffer a cyber breach than the one with the best risk rating. It also confirmed the belief that certain sectors are more vulnerable than others; budget-strapped education entities typically rated poorly while banks scored highly due to heavy investment in security infrastructure.
“Hackers have various tools available to them and are looking for certain vulnerabilities. The scanning software gives the company the same view of its internet footprint as any hacker would get,” explained Newman.
“A company that takes the findings very seriously may well look at every individual asset and negative signal with a view to fixing things and putting in new policies and procedures. That’s where the big work comes in.”
Insurers are also partnering with tech firms to offer similar risk assessment solutions. Allianz recently entered a heavyweight partnership with Aon, Apple and Cisco, harnessing the tech firm’s intelligent Cisco Ransomware Defense software.
Jenny Soubra, head of cyber and tech, Allianz Global Corporate & Specialty, claimed Cisco offers the most holistic cyber security solution in the market, and its counterintelligence measures identified the WannaCry virus a month ahead of the attack, allowing it to warn and protect its customers.
“Different levels of coverage are unlocked by different levels of software and hardware deployment,” Soubra explained, revealing that the insurer is offering users “broad terms and conditions not currently available in the marketplace,” with deductibles in some cases reduced to zero.
Over time, this kind of arrangement is likely to become more prevalent, and machine learning tools will have a big impact on cyber insurance terms and pricing — not only by reducing the risk of insureds suffering breaches but also generating invaluable data to help refine underwriting.
One key challenge: risk aggregation. These tools could help identify vulnerabilities within supply chains, allowing users to suggest security improvements to suppliers and clients.
Costs and Considerations
Uptake is limited. Asked if machine learning is being translated into meaningful risk mitigation, Griffin said, “We haven’t seen that yet on the client side. As an industry, we’re kidding ourselves that we’re going to change the behavior of complex insureds who have invested millions of dollars into an infrastructure and a security team.”
Costs, which vary significantly, may be prohibitive to smaller organizations.
“These programs can be expensive,” said Peters, noting annual prices on a per-user or per-instance basis can range from $1,000 to $20,000 and upwards. “It’s not for the faint of heart.” And while the range of machine learning tools continues to grow and improve, they do not offer a silver-bullet solution: “These platforms are only as good as the person administering them,” Peters said.
“It takes a lot of time and effort to fine-tune them to fit your business needs and day-to-day processes. Increasing network traffic may be statistically interesting, but it rarely represents an attack, and systems that look for generic anomalies can often misclassify a threat. You have to know how to apply machine learning in order for it to reveal true insight,” he added.
Newman agrees that false positive or negative signals are possible, such as spotting vulnerabilities in assets that may be disconnected from key services or are of little value or importance. Barbican’s approach, he said, is to use the signals as the starting point for risk mitigation conversations.
Machine learning is still young, and these tools will only become more accurate and effective. However, warned Griffin, “as technology evolves, so does the threat.”
Sophisticated cyber criminals will soon harness the power of machine learning to make their attacks more effective. This could make smaller firms without intelligent security systems particularly vulnerable — not only for the assets they possess but also as routes into larger organizations.
Cernak hopes machine learning programs will become easier to deploy, improving smaller companies’ access to these powerful tools. However, even the adoption of high-end software is pointless if the human principles of cyber security are not adhered to. Adequately training staff remains vital as human error is almost always present in the event of a breach.
“You have to treat machine learning like any other tool in your toolbox. You can’t become overly dependent on any one solution,” said Cernak. “It’s important to build a culture of risk awareness and prevention. Cyber security starts and ends with people, and the tools you deploy need to be complementary to that strategy.” &