Opinion & Analysis
Written by: Elijah Fernandez | MSW, PMP, CISSP | CEREVITY
Updated 8:00 AM EST, November 17, 2026

The question executives ask before they agree to see a therapist is almost always the same: Is therapy confidential? They put it to the clinician. The honest answer usually belongs to someone else. By the time a clinician makes a promise in the first session, the people who designed the intake form, the scheduling system, the billing flow, and the analytics feed have already decided how much of that promise the organization can keep.
I have spent about 20 years in healthcare IT, including as director of IT operations at a national mental health company, and I now build the technology behind a therapy network for senior leaders. The lesson I keep relearning is that confidentiality is not a clinical virtue that data teams support. It is a data governance outcome that clinicians inherit.
That matters well beyond healthcare. Any chief data officer (CDO) or data leader whose organization asks people to share something they would not say in a meeting, from an employee assistance program to an HR complaint channel to a wellness benefit, is running a trust product. The governance either earns the trust or quietly spends it.
In my experience, leaders can be hesitant to use traditional employee assistance programs or insurance-based care because they’re concerned about where their personal information might end up. I would describe that as a data problem before I would describe it as stigma.
When people do not trust where their information goes, they do not stop needing help. They delay, they use a personal email address, they pay cash somewhere else, or they say less. Every one of those workarounds degrades the data the organization does hold and pushes the real story outside any system it governs. Low trust shows up as missing data long before it shows up in a survey.
HIPAA has a principle I wish more data programs outside healthcare borrowed. The minimum necessary standard in 45 CFR 164.502(b) asks covered entities to make reasonable efforts to limit protected health information to the minimum needed for the intended purpose. It is a legal rule, but it reads like a design rule.
Day to day, the most protective control I know is the field that never gets built. A field that does not exist cannot be breached, subpoenaed, exported to the wrong dashboard, or reused for a purpose nobody approved. So before a new field ships, I want three answers in writing: what decision depends on it, who is allowed to see it, and when it will be deleted. If nobody can name the decision, the field does not ship.
This sounds obvious, but it is rarely enforced because the cost of collecting one more data point is paid later by someone else, usually in an incident review.
Most governance programs catalog where data rests. The exposure in sensitive data usually comes from where it moves: the support tool that logs a conversation, the marketing pixel on a booking page, or the analytics extract built for a one-time question that never got turned off.
Regulators have caught up to this. In April 2024, the Federal Trade Commission finalized changes to its Health Breach Notification Rule, underscoring that it applies to certain health apps and similar technologies not covered by HIPAA and making clear that a breach of security includes an unauthorized disclosure, not only a hack. In other words, an unauthorized disclosure of covered health information can constitute a breach even if no attacker was involved.
For a data leader, that turns the integration inventory into a security document. Every outbound connection from a system that holds sensitive data deserves the same scrutiny as an inbound one.
The architecture decision with the biggest effect on trust is keeping identity and logistics apart from sensitive content. Scheduling and matching systems need to know who a person is and when they are available. Payment systems need the information necessary to process a transaction. These systems don’t need to know why the person is there.
One example from our own stack: in the clinical record system our network uses, secure messages between a clinician and a client are visible only to the two people in that thread. There is no administrative inbox, and the account owner cannot read them. That creates real operational friction. When a client sends a late change through a message, operations does not see it. We accept that cost because a confidentiality promise that has an administrator exception is not the promise people think they are getting.
Data leaders face the same trade every time someone asks for a single view of the customer or the employee. A unified view is valuable. It is also the exact thing a person sharing something sensitive is afraid of.
Most privacy statements are written by legal or marketing first and reconciled with the systems later, if ever. Reverse the order. Draw the data map, including every flow and every person with access, and then write a promise that the map can prove. A smaller promise that is true earns more trust than a sweeping one with footnotes.
Three questions are worth bringing to your next governance review:
The executives who ask whether therapy is confidential are really asking whether the system around the therapist deserves their trust. That answer is written in schemas, access rules, and retention schedules. It belongs to data leaders, and the people deciding whether to ask for help early are reading it closely.