As Good As They're Getting, Analytics Don't Inherently Protect Data
It is only a matter of time before your system is breached, and when your data is lost, analytics won't help you.
“Use analytics to secure your systems!” This phrase is becoming more common in today’s over-hyped security industry. Certainly the tools continue to improve with real-time data collection, large scale manipulation and advanced visualization. And marketing terminology keeps pace with “machine learning,” “predictive analytics,” and “Big Data.” These innovations deliver better insight into complex systems, and there is no question that they play a role in managing networks and resources. But analytics is just a fancy word for monitoring.
The suggestion to “use analytics to secure your system” is flawed, and the argument to shift away from data security systems like encryption and move to analytics is fallacious. In fact, analytics is not an either-or-choice with encryption. Suggesting that firms choose between the two is like a doctor telling a patient to choose either vitamins or exercise. Both have their place in a healthy lifestyle.
As good as they are getting, analytics does not inherently protect data. Therefore it is unequivocally prudent to use encryption.
[COUNTERPOINT: Encryption Has Its Place But It Isn’t Foolproof by Doug Clare, Vice President of Product Management, FICO]
Let’s say a healthcare company processes HIPAA data and experiences a breach. A well-functioning analytics package would be able to tell the incident team how the breach happened and what data was leaked. It might have even alerted teams that they were subject to an attack so they could respond. But if data leaked, the company’s remediation steps are vastly different depending on whether or not they encrypted their data. Pick any victim of one of 2015’s high visibility data breaches - from Anthem to OPM - and re-think the impact if that data had been properly encrypted.
Encryption is designed to work best when unauthorized parties attempt to access data. Encryption is a last line of defense to protect data once the breach has occurred. And be sure about it, breaches will occur.
Not if, but when
Prudent infosec professionals think about security in the context of reducing threat surface area and minimizing damage in the event of the exploit. The joke used to be that the only secure system is the one that isn’t connected to the Internet. But with the spate of embedded systems exploits - like the firmware hack against hard disk controllers -- that joke needed an update. Experts agree: there is no such thing as a completely secure system.
If you share this same perspective as the professionals, then much of your risk assessment must focus on what happens post-breach.
Quantifying your risk is where the true value of encryption comes into play. Your risk profile is determined by the cost of data loss. Whether reputational damage, loss of certifications or accreditations, or other legal clean-up fees, the usefulness of the data to the bad guys will determine the level of risk to your organization. Encryption minimizes this risk by reducing the usefulness of the data. Analytics does not do that.
While encryption is the best and last line of defense in the event of a breach, the sad truth is that it isn’t as mainstream as it should be. The underlying methods are decades old and relatively straightforward to implement, but industry has been unable or unwilling to deploy it.
That isn’t because they’re spending their time monitoring and remediating systems. In fact, most organizations don’t follow the basic processes that would protect them from obvious threats. The Sony Pictures breach from 2014 highlights this perfectly. It was nine months after the breach that the hackers told them they had their data. During that nine-month period, the bad guys poked around the SPE network, installed an egress server in the Sony DMZ, and extracted terabytes of data. Basic system and network monitoring should have flagged these anomalous activities. Why didn’t Sony know?
Similarly, congressional testimony following the OPM breach showed that they’d been informed of the vulnerability of their systems for years, up to and including specific recommendations by the Inspector General, which were summarily ignored.
Organizations don’t do well with repetitive, mundane tasks like monitoring, interpreting, and remediating. It’s kind of like flossing your teeth: even the most committed among us will skip a day or two. With more sophisticated analytics tools generating more real-time data, you need more sophisticated people and processes, which are expensive. That is why outsourcing the SOC is becoming more common.
The human requirement
The analytics side would have you believe that “machine learning” will reduce or eliminate the burden on IT. No technology is so advanced that it is fully autonomous -- and it would be dangerous to assume so. But it can reduce the burden.
Systems need to be trained by humans using data that represents the ecosystem well enough that anomalies can be identified. That’s hard, since 25% of all data breaches are due to human error. Those systems then need to be maintained by humans in order to balance the false-positive/missed-negative tradeoffs, and they’ll need to be re-trained as underlying systems change.
Even the best predictive models still require grey matter analysis. We saw this at my last company, Postini, where our “machine learning system” processed more than three billion email messages per day. Each message would have a set of attributes we’d use to identify legit email separately from spam, phishing, and virus-laden messages. As good as our systems were and as diverse as the training sample was, we still needed humans performing around-the-clock review and adjust in order to stay current.
That was just for email threats. The diversity of signals generated across Web services, data stores, and applications is orders of magnitude more complicated than an email payload. Machines cannot auto-detect and auto-resolve.
From a total cost of ownership perspective, I’d also assert that implementing robust encryption is less expensive and operationally simpler than relying on analytics. Encryption requires specialized skills and a re-factoring of data, data stores, and applications. That work is done by software engineering teams through standard software development processes. It doesn’t require the difficult cultural changes and the 24x7 attention that are necessary to make full use of analytics.
It is only a matter of time before your system is breached, and when your data is lost, analytics won’t help you. Would you prefer that your data was out there in the clear, or encrypted?
About the Author
You May Also Like
A Cyber Pros' Guide to Navigating Emerging Privacy Regulation
Dec 10, 2024Identifying the Cybersecurity Metrics that Actually Matter
Dec 11, 2024The Current State of AI Adoption in Cybersecurity, Including its Opportunities
Dec 12, 2024Cybersecurity Day: How to Automate Security Analytics with AI and ML
Dec 17, 2024The Dirt on ROT Data
Dec 18, 2024