Preventing Data Loss, At Scale
Christian Stegh, CTO and VP of Strategy at eGroup Enabling Technologies.
gettyData loss prevention tools have spiked in popularity since the arrival of enterprise AI. But even though tools exist to protect enterprise data, operating them is far from foolproof. CISOs have good reason to deploy tools to protect their corporate assets. Oversharing and over-permissioned data can lead to leakage, and the average cost of an insider security incident is $19.5 million, dwarfing the average ransomware payment of $139,000. This article outlines lessons learned from six years of enterprise experience rolling out common DLP products like Varonis and Microsoft Purview.
1. Enterprise AI tools like Copilot can search and find any unprotected data within the enterprise to which that user has access.
2. Confidential corporate data may be uploaded to public AI tools, which, unless turned off, use the input to train their models by default.
To reduce these risks, organizations are investing in tools. Those that are succeeding add user adoption and process improvements to the equation.
There are myriad challenges of successfully adopting DLP tools at scale, but four common themes arise.
2. A vague understanding of the risk.
3. Little time or interest from the business to change behavior.
4. Inadequate resources to operate DLP systems at scale.
The first two are closely related. For most organizations, people have created, shared and stored data with little consideration about its life cycle or protection. Relatively few organizations can answer, “How many sensitive documents are stored in which locations?” For some, simply defining what data is sensitive is challenging! Without awareness of the scope of the sprawl, there’s no shared understanding of the severity of the risk, and no sustained action.
No matter the type of business, challenge three, business resistance to change, is ubiquitous. Business leaders rarely see value in interrupting their work for DLP. DLP has been IT’s job.
Speaking of IT’s job, that’s challenge number four, and a common question: Should operating DLP be a full-time job? DLP tools create false positives and need tuning to scale. CISA recommends a maintenance process with “recurring activity with tasks scheduled monthly, semiannually, and annually.” Tasking an already at-capacity SecOps team with more alerts can create a shock to the support team, halting rollouts before they scale.
Use simple frameworks for data classification. Varonis suggests three classes, and Microsoft between three and five classes (i.e., public, internal, confidential, secret).
Getting the business to agree on a framework is easier in organizations where data is sacred. Take for instance, health insurance and benefits provider PlanSource, who implemented four classes of data. Since they administer benefits programs for employees of 1,000 organizations, their CISO, Tony Spurlin, is on high alert. “We are a data company, and what comes with data is a significant amount of trust. You have one chance with trust. You break it once and that’s it.”
It’s challenging to begin the DLP journey by labeling and protecting the organization’s niche, proprietary data, even if it’s truly the most sensitive. Agreeing on how to handle it and who needs access isn’t straightforward.
A simpler step is to start with data everyone agrees should be protected, like employees’ personally identifiable or health information. That’s simpler to scan and protect, and there’s no argument that it needs protection. Next, protect complex data like intellectual property, customer content and unique identifiers.
Auto-labeling sounds great in the marketing slides. So it may sound counterintuitive to suggest that people start by manually classifying their own files. Consider the upsides, though. Manually labeling helps people build their data literacy (solving problem #1 above). Plus, manual labeling is possible with the less expensive Purview license, unlike the upgrade needed to auto-label data. Later, once people understand the process, label all that at-rest data.
There’s no question that user acceptance leads to ROI, but how do you make people care about a dry topic like labeling files? It’s about awareness of the need to change, as well as the tactical how-tos. This philosophy is backed up by SANS, who not only stresses the “why” but how to implement a strong feedback loop. Looping back to suggestion one: By protecting employees’ own information first, you’ll get more support for the larger initiative.
The rollout of any DLP solution is iterative, and vendor marketecture can set some misleading expectations. Microsoft’s Purview deployment guide calls for five weeks of setup but notes that the time to “operate, expand and retroactive options” is “situational.” Gartner Group analyst Mike Huskey said, “Yes, you could ‘deploy’ Purview in 6 weeks, or in an afternoon, if you just want to click buttons.”
To scale anything but the simplest rollout, plan for six-plus months. If you plan ahead and set up data literacy and milestones up front, things can move quickly. If planning, data literacy and milestones are set up front, as done at Verdantas, things can move “This is not a 12-month or 18-month project, but is something that you can do pretty quickly,” said CIO Sujan Turlapaty.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
