NIST’s privacy draft counted harm where customers feel it

NIST’s September 2019 privacy draft measured risk by the problems people experience from data processing, even when nothing is breached. That moved the evidence for privacy risk from security logs to the places customers, patients and citizens meet an organization.

Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.

Two-minute summaryThe dek and the first sentence of each section, quoted from the article

NIST’s September 2019 privacy draft measured risk by the problems people experience from data processing, even when nothing is breached. That moved the evidence for privacy risk from security logs to the places customers, patients and citizens meet an organization.

  1. The week On September 9, 2019, the National Institute of Standards and Technology announced the preliminary draft of the NIST Privacy Framework: A Tool for Improving Privacy through Enterprise Risk Management.
  2. Harm without a breach The draft drew a simple diagram.
  3. The person as the unit of measure The draft’s most consequential sentence may be its simplest.
  4. What people are told and what they can do Two of the draft’s functions dealt directly with the person’s side of the relationship.
  5. The partners in the processing Customer data in a regulated organization can pass through many other organizations: cloud providers, core processors, electronic health record vendors, payment networks, marketing platforms, analytics services, contact center providers and, for agencies, contractors who run whole programs.
  6. Where the argument is weakest The argument here is that the draft made the person’s experience the measure of privacy risk, and so moved the evidence toward customer-facing teams.
  7. With hindsight NIST released version 1.0 of the Privacy Framework on January 16, 2020.

The week

On September 9, 2019, the National Institute of Standards and Technology announced the preliminary draft of the NIST Privacy Framework: A Tool for Improving Privacy through Enterprise Risk Management. The draft itself was dated September 6. The same day as the announcement, a notice in the Federal Register, at 84 FR 47255, asked for public comments by 5:00 p.m. Eastern time on October 24, 2019. NIST said it would post the comments it received without change or redaction, and that it would use them to develop version 1.0.

The draft followed public work that the notice described. NIST had published a request for information in the Federal Register on November 14, 2018. It held three public workshops: in Austin in October 2018, in Atlanta in May 2019 and in Boise in July 2019. It held four webinars and released an outline in February 2019, a discussion draft in April and supplemental materials in June.

The framework was voluntary. NIST modeled its structure on the Framework for Improving Critical Infrastructure Cybersecurity, version 1.1, so that organizations could use the two together. It had three parts. The Core set out activities and outcomes in five functions: Identify-P, Govern-P, Control-P, Communicate-P and Protect-P. The “-P” marked them as privacy functions, distinct from the cybersecurity framework’s. Profiles let an organization pick the outcomes that mattered most to it and compare a current state with a target. Implementation Tiers described how mature its privacy risk management was.

Each function broke down into categories. Identify-P covered inventory and mapping, the business environment, risk assessment, and risk management for the data processing ecosystem. Govern-P covered policies and procedures, risk management strategy, awareness and training, and monitoring and review. Control-P covered data management policies, data management and what NIST called disassociated processing. Communicate-P covered communication policies and data processing awareness. Protect-P covered identity management, authentication and access control, data security, data protection policies, maintenance and protective technology. The draft also included a set of privacy risk management practices and a placeholder for a roadmap of areas where practice was not yet well understood.

The draft set out a particular view of privacy risk. NIST said it considers the problems individuals could experience from a system’s, product’s or service’s operations with data, across the full life cycle from collection to disposal. It called those operations data actions, and their sum data processing. The problems ranged from dignity-type effects such as embarrassment or stigma to more tangible harms such as discrimination, economic loss or physical harm. Problems could arise as unintended consequences of processing an organization carried out to meet its own mission or business objectives. The first four functions addressed risks arising from data processing; Protect-P could help manage risks from privacy breaches.

NIST asked reviewers specific questions. Among them were whether the draft adequately defined the relationship between privacy and cybersecurity risk, whether it appropriately integrated privacy risk into organizational risk, and whether it let people in different roles, from boards and senior executives to legal, compliance, security and IT staff, understand privacy risks at the right level of detail. A note to reviewers said a repository and a process for accepting external informative references would come with version 1.0.

NIST’s privacy draft followed the structure of its cybersecurity framework, and in the draft’s own diagram cybersecurity risk is associated with loss of confidentiality, integrity or availability. The privacy draft measured something else. Its unit of risk was a problem experienced by a person, and it said such problems can arise when nothing has been breached. That choice moved the evidence for privacy risk out of the security team’s logs and into the places where customers, patients and citizens meet an organization’s systems. This article takes the argument in five parts: harm without a breach, the person as the unit of measure, what people are told and can do, the partners in the processing, and where the argument is weakest.

Harm without a breach

The draft drew a simple diagram. One circle held cybersecurity risks, associated with loss of confidentiality, integrity or availability. The other held privacy risks, associated with unintended consequences of data processing. The overlap was the privacy breach. Everything else in the privacy circle was risk that a security program, however strong, would not see.

The smart meter example

NIST illustrated the point with smart meters installed as part of the Smart Grid. The meters could collect, record and distribute granular information about household electricity use, which could give insight into people’s behavior inside their homes. The meters were operating as intended. The data processing could still leave people feeling surveilled. No attacker, no misconfiguration and no lost laptop was needed for the problem to occur.

The example generalizes to every regulated sector. A bank’s transaction data can reveal where a customer shops, worships or receives medical care. A health system’s appointment and portal data can reveal conditions a patient has not told family or an employer. A public agency’s benefits records can reveal income, disability and household changes. In each case the organization may hold the data lawfully, protect it well and still use it in ways the person did not expect.

Problematic data actions

The draft gave this kind of risk a name. When an organization can identify the likelihood that a data action will cause a problem for individuals, NIST called it a problematic data action. The organization can then assess the impact should that action occur. The Core’s risk assessment category asked organizations to identify contextual factors, including individuals’ demographics and privacy interests or perceptions, data sensitivity, and how visible the processing is to individuals and third parties. It asked them to identify potential problematic data actions and to evaluate data analytic inputs and outputs for bias.

Identifying a problematic data action requires no incident. It requires knowing what each system does with data and why. The draft’s inventory and mapping category asked organizations to inventory systems, the parties that own or operate them, the categories of individuals whose data they process, the data actions, the purposes and the data elements. Those are design facts. They live with the product owners, architects and business teams who decide what a system collects and how it uses it.

Purpose as the test

The draft’s inventory asked for the purposes of each data action, and its disassociated processing category included an outcome that processing is limited to what is relevant and necessary for a system to meet mission or business objectives. Purpose is where a person’s expectations and an organization’s plans meet. A customer who gives an address to receive statements expects it to be used for statements. A later decision to use it for marketing, partner offers or analytics adds a purpose the person never saw.

Recording purpose at the level of each data action, and requiring a fresh review when a purpose is added, turns purpose from a line in a privacy notice into a control. The review asks whether the new purpose would surprise the people affected. If it would, the organization can decide to tell them, give them a choice, change the design or not proceed.

Data actions across the life cycle

The draft defined a data action broadly: any operation in a system’s data life cycle, including collection, retention, logging, generation, transformation, use, disclosure, sharing, transmission and disposal. Several of those verbs can fall outside a privacy review that looks only at collection and sharing. Logging and retention can be treated as operational settings. Generation covers data an organization creates about a person, such as a score, a segment or an inference. Transformation covers joins and enrichment.

Each is a point where a person can be affected without any data leaving the organization. A retention period longer than the purpose needs keeps a record available for uses no one has yet decided on. A model that generates a propensity score creates a new fact about a customer that the customer never provided. An inventory that lists only collection and disclosure misses most of the life cycle the draft described.

Why security programs miss it

A security program looks for deviations: unauthorized access, unexpected traffic, altered records. A problematic data action is authorized by definition. A marketing team that joins transaction data to a vendor’s demographic file, an analytics script that records page views in a patient portal, or an eligibility model that infers household changes from new data sources are all doing what someone approved. The security tools would record them as normal. The draft’s point was that normal is exactly where privacy risk can sit, and that a second line of sight is needed to find it.

That second line of sight has to come from somewhere. Change management, product approval and procurement are the moments when a new data action is decided. A privacy review that runs at those moments sees the risk before it reaches customers. A review that runs only after an incident sees the subset of privacy risk that overlaps with security.

Security tools are data actions too

The security program itself processes personal data. Logs record who did what and when. Monitoring tools inspect traffic and content. Insider-threat analytics profile employee behavior. Fraud systems score customers’ transactions and devices. The draft’s data management category included an outcome that audit and log records are determined, documented, implemented and reviewed in accordance with policy and incorporating the principle of data minimization.

That makes security a subject of privacy risk assessment as well as a control for it. A bank’s fraud model that collects device and location signals, a health system’s access-monitoring tool that records every chart a clinician opens, and an agency’s identity proofing service that checks applicants against outside data sources all affect people. Each is justified. Each still has a purpose, data elements, a retention period and a set of people it touches, and the person-centered method applies to it.

The person as the unit of measure

The draft’s most consequential sentence may be its simplest. Individuals, singly or in groups, including at a societal level, experience the direct impact of problems. The organization experiences impacts as a result: noncompliance costs, customers abandoning products and services, harm to its reputation or internal culture. NIST described the impact assessment as the point where privacy risk and organizational risk intersect.

Two steps, in a fixed order

That ordering changes the risk register. A conventional enterprise risk entry starts with the organization: what could we lose, and how likely is it. The privacy draft asked organizations to start with the person: what could happen to them, how likely is it, and how bad would it be for them. Only then does the organization translate that into its own exposure. The draft said those organizational impacts could drive decisions about resources and help bring privacy risk into parity with other enterprise risks.

The difference shows up in prioritization. Measured by organizational loss, a data action that affects a small group of vulnerable customers can look minor. Measured by the problems those customers would experience, it may rank near the top. A bank deciding whether to use transaction data to flag customers for hardship outreach, a health system deciding what appointment reminders reveal on a lock screen, and an agency deciding what a benefits notice says in a shared mailbox are all decisions where the person’s experience is the measure.

Translating harm into enterprise terms

The draft did not ask risk committees to abandon their usual measures. It asked them to derive those measures from the person’s problem. A data action that would embarrass customers translates into complaints, attrition and reputational damage. One that would cause economic loss translates into remediation, restitution and regulatory exposure. One that discriminates translates into legal liability and loss of trust in a community. Writing the person’s problem first, and the enterprise consequence second, keeps the chain of reasoning visible to the people who approve the risk.

It also makes risk acceptance more honest. A committee that accepts a residual risk is accepting that some number of people may experience a specific problem. Stating the problem in those terms, rather than as an abstract rating, gives the decision its actual content.

Groups as well as individuals

The draft said problems can be experienced by individuals singly or in groups, including at a societal level. That clause reaches risks a record-by-record review would miss. Aggregate data can harm a group even when no individual is named. A lender’s analysis that marks certain neighborhoods as higher risk, a health system’s published statistics on a small community, or an agency’s open data release broken down to small geographic units can stigmatize the people in them or expose them to unwanted inference. De-identification is meant to protect individuals from being singled out. It does not, on its own, protect a group from what the data says about it. A risk assessment that asks who could be affected as a group, and how, covers that gap.

Knowing who the people are

Measuring harm to individuals requires knowing which individuals are involved. The draft’s Core asked organizations to inventory the categories of individuals whose data they process, such as customers, employees, prospective employees and consumers. Regulated organizations serve populations with very different expectations and vulnerabilities. A credit union’s members, a wealth manager’s clients, a hospital’s patients, a behavioral health clinic’s patients and a state agency’s benefit recipients would not experience the same data action in the same way.

That is a customer and citizen experience question before it is a compliance question. The teams that research users, design journeys, handle complaints and run call centers can already hold evidence about what people expect. The draft’s risk assessment category named individuals’ privacy interests and perceptions as a contextual factor. The natural source of that evidence is the organization’s own record of what customers say.

Inference and bias

The draft’s risk assessment category included an outcome that data analytic inputs and outputs are identified and evaluated for bias. Placing bias inside privacy risk assessment linked the two. A model that infers something about a person, such as a likelihood of default, a health risk or eligibility for a program, generates data the person did not provide and may not know exists. If the inference is wrong or skewed for a group, the problem falls on the people in that group.

For regulated organizations, those models can carry other obligations too, under lending, health or civil rights law. The privacy lens adds a question those regimes may not ask: whether the people affected would expect the inference at all, and whether they can see or contest it. That question is answered by asking them, testing explanations and watching what they do when told.

Where the evidence sits

If the person is the unit of measure, the evidence of privacy risk is in places a security team may not read. Complaints that a notice was confusing. Call center contacts asking why an offer reflected a recent purchase. Portal feedback about what a caregiver could see. Opt-out rates after a new data use. Requests to delete or correct data. The draft’s monitoring category included processes for receiving, tracking and responding to complaints, concerns and questions from individuals about privacy practices, and for incorporating lessons learned from problematic data actions.

A practical program would route those signals to whoever owns privacy risk. A complaint taxonomy with privacy categories, a monthly review of privacy-related contacts by product, and a threshold that triggers a fresh risk assessment would turn customer experience data into privacy risk data. Most of these signals come from channels the organization runs itself. The change is in reading them as evidence.

A starting inventory

An organization starting from nothing does not need a complete map before it acts. The draft’s profile mechanism allowed an organization to choose the outcomes most important to it. A sensible first scope is the set of systems that touch customers directly: websites, apps, portals, contact center platforms and marketing systems. For each, the inventory records purposes, data elements, parties, retention and the categories of people affected. Those are the systems where a problematic data action would be felt soonest, and where customer evidence is easiest to gather.

What people are told and what they can do

Two of the draft’s functions dealt directly with the person’s side of the relationship. Communicate-P covered what individuals and organizations know about data processing. Control-P covered whether data can be managed with enough granularity to manage privacy risk, including individuals’ own preferences and requests.

Communication as a control

The draft treated communication as a risk control, not as a legal formality. Its data processing awareness category described mechanisms such as notices and internal or public reports for communicating purposes, practices, risks and options for individuals’ preferences and requests. It included mechanisms for obtaining feedback from individuals, such as surveys or focus groups. It called for system and product design that enables data processing visibility. And it called for mitigation mechanisms to address impacts on individuals.

Each of those is a design artifact. A notice is a piece of content in a digital journey. A preference center is a feature. A just-in-time explanation of why an app asks for location is an interface decision. The people who build them are product managers, content designers and engineers. If communication is a control, those teams operate it, and the privacy function’s job is to set the standard and test whether people actually understand.

Visibility by design

Two of the draft’s communication outcomes concerned what a system itself shows. One said system, product and service design enables data processing visibility. Another said records of data disclosures and sharing are maintained and can be accessed for review or transmission. Together they point to features rather than documents: an account page that lists which partners received a customer’s data and when, a portal view that shows who accessed a patient’s record, or a benefits account that shows which agencies a resident’s information was shared with. Each depends on disclosure records that the back-end systems keep in a form the front end can display.

Testing whether people understand

The framework’s inclusion of surveys and focus groups as communication mechanisms implies a measurable outcome: individuals have reliable knowledge about data processing. That can be tested the way any customer experience is tested. Comprehension testing of notices with real users, task completion rates in preference centers, and the share of customers who can correctly say what a feature does with their data are all measurable. A privacy program that reports those measures to its risk committee is reporting on the outcome the draft described. One that reports only the number of notices published is reporting on activity.

Control as a capability

Control-P’s data management category described capabilities: data elements that can be accessed for review, transmission, alteration and deletion; data destroyed according to policy; metadata carrying processing permissions transmitted with the data. Its policy category called for processes to authorize data processing, including through individual consent, to revoke authorizations, and to enable individuals’ data processing preferences and requests.

Those are architecture requirements as much as policies. Honoring a customer’s choice means the choice has to travel with the data to every system that uses it. A consent recorded in a marketing platform does nothing if the analytics warehouse never sees it. A correction made in the core banking system does nothing for the copy a vendor already holds. The draft’s communication category called for corrections or deletions to be communicable to other parties in the ecosystem and for records of disclosures, provenance and lineage to be maintained.

Requests, mitigation and redress

The draft treated individuals’ requests as a designed capability. Its policy category called for processes enabling individuals’ data processing preferences and requests. Its awareness category called for mitigation mechanisms to address impacts on individuals that arise from data processing. Its monitoring category called for receiving, tracking and responding to complaints, concerns and questions.

Together those describe a service, with intake, routing, fulfillment and closure. A regulated organization would measure it like any other service: time to acknowledge, time to complete, rework, escalations and the reasons people give. Redress deserves its own design. When a data action has already caused a problem, the person needs a way to have it corrected and, where appropriate, a remedy. A process that can only answer “what data do you hold about me” is not yet a mitigation mechanism.

Notifying about events as well as breaches

One outcome in the draft read: impacted individuals and organizations are notified about a privacy breach or event. The words “or event” widened a familiar obligation. Breach notification is a legal duty in many settings, with its own triggers and timelines. The draft’s outcome suggested telling people about problems from authorized processing as well, such as a data use that went further than they were told or an inference that proved wrong. Whether and how to do that is a judgment, and it may carry legal consequences that counsel should weigh. The draft put it on the list of outcomes an organization might choose.

Disassociated processing

The draft also named a group of technical outcomes it called disassociated processing. They included processing data in unobservable or unlinkable ways, limiting identification through techniques such as differential privacy and tokenization, restricting inferences about individuals’ behavior, permitting selective collection, and limiting processing to what is relevant and necessary. For a customer, these are invisible when they work. They reduce the number of problems that communication and control would otherwise have to manage. An organization that tokenizes identifiers before analytics, or computes a result on a device instead of collecting raw data, has fewer data actions to explain.

The partners in the processing

Customer data in a regulated organization can pass through many other organizations: cloud providers, core processors, electronic health record vendors, payment networks, marketing platforms, analytics services, contact center providers and, for agencies, contractors who run whole programs. The draft devoted a category and a section to this data processing ecosystem.

Ecosystem risk management

The ecosystem category asked organizations to identify, prioritize and assess ecosystem parties using a privacy risk assessment process; to use contracts to implement measures that meet the privacy program’s objectives; and to assess those parties routinely through audits, test results or other evaluations. The section on the ecosystem noted that relationships can be complex and multi-directional, with service providers supported by further service providers. It said an organization should manage privacy risk from its own standpoint and consider how its choices affect other parties’ management of privacy risk.

For the customer, the ecosystem is invisible. A patient who uses a portal sees the health system’s brand, not the analytics script inside it. A bank customer sees the bank’s app, not the aggregator or marketing vendor behind a feature. When a partner’s data action causes a problem, the person experiences it as something the organization they chose did to them. The draft’s customer-centered measure therefore extends to every partner in the chain.

Profiles as purchasing requirements

The draft suggested a concrete use. An organization’s current or target profile can generate a prioritized list of privacy requirements, and those can inform decisions about buying products and services. NIST’s example was a device bought to monitor a forest, where the ability to limit data collected about people using the forest should drive the evaluation. Where it is not possible to impose requirements on a supplier, the draft said the goal should be the best buying decision among suppliers, given a carefully determined list of requirements, with residual risk addressed through mitigation.

Translated to a regulated buyer, that means writing privacy outcomes into requests for proposal: what data the product collects by default, whether collection can be turned off by element, whether consent and preference signals pass through, whether corrections and deletions propagate, and what records of disclosure the vendor keeps. Those are questions a procurement team can ask and a vendor can answer in writing. They connect the customer’s experience to a contract term.

Organizations with more than one role

The draft noted that an organization may play several roles at once, such as providing services to other organizations while selling products to consumers. It observed that some laws classify organizations as data controllers or data processors, and that roles can also come from industry designations. It told organizations to use the framework from their own standpoint in the ecosystem.

A regulated organization can wear more than one hat. A bank that offers accounts to its own customers may also process payments for fintech partners whose customers it never sees. A health system that treats its own patients may also host records or run technology for affiliated practices. A state agency that serves residents directly may also supply data to federal partners or to other state programs. In each case, the same data can be both the organization’s own customer data and data it holds for someone else. The person-centered view helps here, because the person does not change with the contract. The risk assessment can follow the person across roles, while the contracts allocate who answers for which data action.

Assessing partners after signing

The ecosystem category ended with an outcome that parties are routinely assessed using audits, test results or other evaluations to confirm they meet their contractual or framework obligations. That is the step contracts alone cannot provide. A clause promising that consent signals will be honored is a claim. A test that sends a withdrawal and checks the partner’s system is evidence. Periodic tests of that kind, focused on the data actions that matter most to customers, give an organization something to show when a person asks what happened to their data.

Shared frameworks among partners

The ecosystem category also included an outcome that interoperability frameworks or similar multi-party approaches are used to manage ecosystem privacy risks. That recognizes a limit of bilateral contracts. When many organizations exchange the same kinds of data, a common set of terms, consent signals and technical standards can carry a person’s choices further than any single contract. For regulated sectors that already exchange data through networks and shared standards, the question is whether those arrangements carry privacy preferences and corrections as reliably as they carry the data itself.

Tags, scripts and embedded services

The hardest ecosystem parties to see are the ones embedded in an organization’s own digital channels. Website and app components supplied by third parties can receive data about what a person does on a page. The organization chose to deploy them. If a marketing or web team added them without the vendor review a core system receives, they are missing from the inventory too. Under the draft’s model, each one is a data action with a purpose, a set of data elements and a party, and belongs in the inventory. A periodic scan of production pages and apps for third-party components, reconciled against the approved inventory, is one way to find the gaps.

Where the argument is weakest

The argument here is that the draft made the person’s experience the measure of privacy risk, and so moved the evidence toward customer-facing teams. Several objections deserve an answer.

The framework was voluntary

The draft was for voluntary use. Regulated organizations already answer to sector privacy laws, examiners and auditors, and a voluntary framework competes for attention with those obligations. That is a fair point about priority. It is weaker as a point about method. The draft did not ask organizations to do something instead of compliance. It offered a way to find risks that compliance checklists can miss, and NIST asked reviewers whether it would help organizations meet applicable laws and regulations. An organization can use the person-centered method inside a compliance program without adopting the framework wholesale.

Security still causes most of the visible harm

A second objection is that breaches, not authorized processing, produce the harms that reach regulators and the public. The draft did not dispute that breaches matter. Protect-P and the overlap with the cybersecurity framework covered them. Its argument was that a program focused only on breaches would miss the other part of the circle. The two views are additive. The risk is in assuming that a strong security program has covered privacy.

Customer evidence is noisy

Complaints and feedback are incomplete and biased toward people who speak up. A person who never notices a data action does not complain about it, even if it would trouble them. That limits customer evidence as a measure, and it is why the draft paired it with inventories, mapping and structured risk assessment. Customer signals tell an organization where people are already surprised. The inventory tells it where they would be if they knew.

Ownership can blur

Moving evidence toward product and customer teams risks diluting accountability. If everyone owns privacy, no one does. The draft’s governance category addressed roles and responsibilities, and its note to reviewers asked whether different roles could understand privacy risk at the right level. A workable split keeps a named privacy owner accountable for the method and the risk register, while making product owners accountable for the inventory, assessments and customer measures of their own products.

Tiers can become the goal

The draft’s Implementation Tiers described a progression from informal, reactive responses to agile, risk-informed approaches. Tiers are useful for communicating about process and resources. They carry a risk that any maturity scale carries: an organization can raise its tier without changing what customers experience. A program that reports its tier to the board, but not the comprehension, complaint and request measures that show how people are affected, has measured the organization again. The draft’s own logic points the other way. The tier is a means; the person’s experience is the result.

Small organizations lack the staff

A small credit union, a rural clinic or a county office may have no privacy engineer and no user research team. The person-centered method can look like a large-organization luxury. The draft anticipated the concern: NIST asked reviewers whether it gave organizations of all sizes enough guidance and flexibility, and its profile mechanism let an organization choose a narrow set of outcomes. A small organization can start with its customer-facing systems, its complaint log and its handful of key vendors. The measure stays the same at any size: what the people it serves would experience.

Differences by kind of organization

For a bank or credit union, the person-centered view bears most on secondary uses of transaction and account data: marketing, profiling and data shared with partners. For a health system, it bears on digital front doors, portals, apps and the third-party components inside them, and on what reminders and messages reveal. For a public agency, it bears on the gap between what a program needs and what its forms and identity checks collect, and on people who have no alternative provider to switch to. A citizen who cannot walk away from a benefits program experiences a problematic data action with fewer options than a customer who can change banks. In each case the specific obligations differ, and organizations should check with counsel on what applies. The method is the same.

With hindsight

NIST released version 1.0 of the Privacy Framework on January 16, 2020. The document states that its contents do not have the force and effect of law. The person-centered approach survived intact. Version 1.0 kept the description of problems ranging from dignity-type effects to discrimination, economic loss and physical harm, and kept the smart meter example. It introduced the term privacy events for the potential problems individuals could experience from data processing. It added revenue loss to the organizational impacts that follow from customers abandoning products and services. And it said Protect-P was specifically focused on cybersecurity-related privacy events, such as privacy breaches. Some category names changed: data management became data processing management, for example.

NIST also built privacy into its federal control catalog. NIST published Special Publication 800-53 Revision 5, Security and Privacy Controls for Information Systems and Organizations, in September 2020, with updates as of December 10, 2020. Its abstract lists privacy risks among the threats the controls address. NIST provides a mapping of the privacy controls in Appendix J of Revision 4 to the integrated control catalog in Revision 5, and crosswalks to the Cybersecurity Framework and the Privacy Framework.

The framework itself moved slowly. NIST began work on version 1.1 to realign it with Cybersecurity Framework 2.0, published a concept paper on June 18, 2024, held a workshop on June 25 and 26, 2024, and released an initial public draft on April 14, 2025. The comment period closed on June 13, 2025. As checked on November 14, 2026, NIST’s page still listed version 1.1 as coming soon. NIST described version 1.1 as a modest update that responds to current privacy risk management needs, realigns with the Cybersecurity Framework 2.0 and improves usability, and listed a separate data governance and management profile among its new projects.

The closest echo of the draft’s view that we found came from a regulator. On July 24, 2025, the California Privacy Protection Agency’s board adopted regulations requiring certain businesses to conduct risk assessments and annual cybersecurity audits, among other changes. The Office of Administrative Law approved them on September 22, 2025, and they took effect on January 1, 2026. A business whose processing presents significant risk to consumers’ privacy, such as selling or sharing personal information or processing sensitive personal information, must assess it before starting. The stated goal is to restrict or prohibit processing when the risks to consumers’ privacy outweigh the benefits.

The required report reads like the 2019 draft turned into rules. It must document how the business interacts with the consumers involved, what disclosures it has made to them and how, and the negative impacts to consumers’ privacy, including impairing their control through insufficient information. How far those rules reach a particular bank, health system or agency depends on the law’s scope, which is a question for counsel. What held up was the method: privacy risk measured by what happens to the person, with the customer relationship as evidence. What did not happen was a quick update of NIST’s own framework.

Spotted an error? Tell us and we will check it and correct the article where needed.

  • NIST Privacy Framework
  • privacy risk
  • data processing
  • consent
  • third-party risk