Chapter 11 · Part 3: Building

Trust by Design: Security, Privacy and Compliance

“Security is a process, not a product.”

— Bruce Schneier

The security questionnaire

Brindlepath’s largest deal of the year nearly died in a spreadsheet. The health system’s information security team sent a questionnaire with 312 questions: how data was encrypted, who could access it, how access was reviewed, how incidents were handled, where backups lived, which subcontractors touched patient data, and how the company knew its code was free of known vulnerabilities.

The sales team had expected the questionnaire to take a week. It took seven, because many of the answers did not exist yet. Engineering knew the data was encrypted, but nobody had documented how. Access reviews happened, but informally. The incident response plan was a page on the company wiki that had not been tested. Maya spent her evenings that month answering questions she wished the product had been designed to answer from the start.

The deal closed. Afterward, Maya made a change that outlasted any feature she shipped that year: security, privacy, compliance and accessibility became part of how every product decision was made, not a review at the end. Customers noticed. The next questionnaire took four days.

Trust is a product feature

The first edition of this book treated security and compliance mostly as regulatory obligations. This edition treats them as the fourth pillar of product leadership from Chapter 1: trust and responsibility. Customers buy software they trust. In healthcare, a breach, a privacy failure or a product that excludes people with disabilities is not just a technical failure; it is a failure of the product, and of the people who lead it.

The cost of getting it wrong is high and rising. IBM’s annual study of data breaches found that in 2025 the average breach cost $4.44 million across industries, and $7.42 million in healthcare, the most expensive industry for the fourteenth year running. Healthcare breaches also took longer to identify and contain: 279 days on average, against 241 across all industries.1

Those numbers measure only the direct costs. They do not capture the deals that never close because a prospect does not trust you, or the customers who quietly leave after an incident.

The landscape of obligations

Every industry has its own rules. The specifics change by country and over time, so treat this section as a map, not legal advice, and work with qualified counsel and compliance professionals.

In US healthcare, a product leader needs to understand at least these:

  • HIPAA and its Privacy, Security and Breach Notification Rules govern protected health information. Software vendors that handle it on behalf of providers or health plans are typically business associates and must sign a business associate agreement (BAA) with each customer, and with their own subcontractors that handle the data.
  • Interoperability and information-blocking rules from the Office of the National Coordinator for Health IT shape how health data must be shared and accessed, and affect products that connect to electronic health records.
  • FDA oversight of software applies when software functions as a medical device. Some clinical decision support is excluded from device regulation when it meets specific criteria, including that clinicians can independently review the basis for its recommendations. Features that diagnose, treat or drive clinical decisions need careful regulatory analysis early.
  • State privacy laws increasingly add obligations for health and consumer data beyond HIPAA.
  • Accessibility requirements are tightening: in 2024, the Department of Health and Human Services finalized a rule requiring organizations that receive HHS funding to make their websites and mobile apps conform to the WCAG 2.1 Level AA standard, with compliance dates phased in from 2026.2

Beyond healthcare, most software sold to businesses must show its security practices through independent assessments such as SOC 2 reports, ISO 27001 certification or, in healthcare, HITRUST. Products sold in Europe must meet the GDPR, and AI features increasingly fall under new rules such as the EU AI Act, whose obligations are phasing in through 2027.

A product leader does not need to be a lawyer. They do need to know which rules apply, which product decisions trigger them, and when to bring in experts.

Security by design

Security by design means building security into every stage of the product’s life, rather than testing for it at the end. For a product leader, that means making sure the following happen as a matter of routine.

Threat modeling at design time. For every significant feature, ask: what data does it touch, who should be able to see or change it, and how could it be abused? Fifteen minutes with an engineer and a whiteboard catches problems that weeks of testing would miss.

Least privilege by default. Users, services and AI agents get only the access they need. Brindlepath’s navigators see only the patients assigned to their team; support staff can view a customer’s configuration but not patient records without an approved, logged request.

Strong identity. Single sign-on with customers’ own identity providers, multi-factor authentication, and prompt removal of access when people leave.

Encryption and audit. Data encrypted in transit and at rest, and every access to sensitive records logged in a way customers can review.

Secure development. Code review, dependency scanning, secrets management and regular penetration testing, built into the delivery flow from Chapter 9.

A tested incident plan. Know who does what when something goes wrong, how customers will be told, and how regulators will be notified. Then practice it.

Privacy by design

Privacy by design, a set of principles developed by Ann Cavoukian, starts from a simple premise: protect personal information by default, not as an afterthought.3 For a product team, it comes down to a few habits.

  • Collect the minimum. Every piece of personal data you collect is a liability as well as an asset. Ask whether you need it before you collect it.
  • Use data only for its stated purpose. Data collected to coordinate a patient’s care should not quietly become training data for an unrelated model.
  • Default to private. The most protective setting should be the default.
  • Be transparent. Tell customers, and where relevant patients, what you collect, why, and who you share it with, in plain language.
  • Plan for deletion. Know how long you keep data, and how you delete or return it at the end of a contract or a product’s life, as Chapter 7 discussed.

A privacy impact assessment for any feature that uses personal data in a new way makes these habits concrete. It need not be long: what data, for what purpose, with what risks, and what mitigations.

Accessibility is part of quality

Accessible products work for people with disabilities: people who are blind or have low vision, who are deaf or hard of hearing, who have motor or cognitive disabilities, or who rely on assistive technology. In healthcare, those users include patients, family caregivers and many clinicians.

Most software still falls short. WebAIM’s 2025 analysis of the home pages of the top one million websites found detectable accessibility failures on 94.8 percent of them, an average of 51 errors per page, most of them common and fixable problems such as low-contrast text and images without text alternatives.4

The international standard is the Web Content Accessibility Guidelines (WCAG), published by the World Wide Web Consortium. Level AA of WCAG 2.1 or 2.2 is the common target. For product teams, accessibility is not a separate project. It is part of the definition of done:

  • Use a design system whose components are accessible by default.
  • Check color contrast, keyboard navigation, screen-reader labels and focus order for every new screen.
  • Test with assistive technology and, where possible, with users who rely on it.
  • Write in plain language. It helps everyone, including people reading under stress.

Accessibility also makes the product better for everyone. Brindlepath’s larger text and keyboard shortcuts, added for navigators with low vision, turned out to be popular with every navigator working a busy afternoon with a headset on.

Working with compliance as a partner

In many companies, compliance is the department that says no at the end of a project. That is usually because it was invited only at the end.

Invite compliance, security and, in healthcare, clinical safety partners into discovery. Share the product brief from Chapter 9 with them. Ask, early, which features will need their review and what they will need to see. Most compliance professionals would far rather shape a design than block a launch.

Build reusable answers. The 312-question security questionnaire taught Brindlepath to maintain a living trust package: security documentation, architecture diagrams, policies, audit reports and standard answers, kept current and shared with prospects early. Some companies publish a public trust center with much of this material. It shortens sales cycles and signals that you take trust seriously.

Brindlepath in practice: the trust package

After the seven-week security questionnaire, Maya asked a product manager to treat the company’s security and compliance documentation as a product, with customers (prospects’ security teams), a backlog and a release rhythm.

The first version of the trust package took six weeks to assemble. It included an architecture overview showing where patient data lived and how it moved; the company’s security policies; the results of its most recent penetration test and its independent security audit; a description of access controls, encryption, logging and backups; the incident response plan; a list of subprocessors that handled customer data, with their roles; the standard business associate agreement; an accessibility conformance report against WCAG 2.1 Level AA; and a library of answers to the hundred most common questionnaire questions, each with an owner and a review date.

Every new questionnaire became input to the backlog. Questions the package could not answer were added, and the gaps they exposed, such as an access review that had happened informally but was not documented, were fixed in the product or the process rather than papered over in an answer. When Brindlepath began using AI models, the package gained a section describing which models were used, for what, with what data, under what agreements, and with what evaluation and monitoring.

The results showed up in the sales cycle. The average time to complete a security review fell from about seven weeks to under two. Prospects’ security teams began telling Brindlepath’s sales team that the package was among the clearest they had seen, which, in a market where trust is everything, became a reason to buy. The work cost a fraction of one product manager’s time each quarter to maintain.

Common mistakes

Treating compliance as a final gate. Reviews at the end can only approve or block. Involving security, privacy and compliance early lets them shape a better design.

Collecting data just in case. Every field of personal data you store is a liability in a breach. Collect what you need, for a stated purpose.

Assuming the vendor handles it. Cloud providers and AI model providers secure their platforms, not your use of them. Responsibility for how your product handles data stays with you.

Leaving accessibility to a later release. Retrofitting accessibility costs far more than building it in, and excludes users in the meantime.

An incident plan nobody has practiced. The middle of a breach is the worst time to discover that nobody knows who calls the customer.

Overclaiming. Marketing claims about security, compliance or clinical outcomes that the product cannot support create legal and reputational risk.

Try this

Pick your product’s most sensitive workflow and walk through it with an engineer, a designer and, if you have one, a security or compliance partner. For each step, ask four questions: What personal data is involved? Who can see or change it, and should they? Is every access logged? Can someone using a screen reader or only a keyboard complete it? Write down every gap you find and add the fixes to your roadmap’s capacity for security and compliance. Then schedule a one-hour incident drill for the same workflow.

How AI changes this

AI features raise new trust questions, and product leaders own the answers.

Data flows multiply. Sending patient data to a third-party model provider is a new data flow with its own agreements, risks and obligations. In healthcare, the provider must typically sign a BAA, commit not to retain or train on the data, and meet your security standards. Map every AI data flow as carefully as any other.

New attack surfaces appear. AI systems can be manipulated through their inputs, for example by hidden instructions inside a document an agent reads, a technique known as prompt injection. Agents that can take actions need strict limits on what they can access and do, and human approval for anything consequential.

Outputs can be wrong, biased or unexplainable. A summary that invents a medication, a risk score that disadvantages a group of patients, or a recommendation nobody can explain are trust failures. Evaluate AI features for accuracy and fairness before launch and monitor them after, as Chapter 17 describes. Frameworks such as the NIST AI Risk Management Framework offer a structured way to think about these risks.5

Transparency matters more. Tell users when content was generated by AI, show the sources behind it where possible, and make it easy for them to correct it. In clinical settings, make clear that AI output supports, and does not replace, professional judgment.

AI can also strengthen trust work. Assistants can draft answers to security questionnaires from your trust package, check new screens for common accessibility failures, review code for security issues, and summarize regulatory changes. As always, an accountable person signs off.

Chapter checklist