OPM’s 2015 fingerprint count rose to 5.6 million, and NIST later wrote that biometrics are not secrets
On September 23, 2015, OPM raised its estimate of people whose fingerprints were stolen to 5.6 million and offered study and future notice. NIST’s 2017 and current identity guidance treat biometrics as non-secret and require a physical authenticator.
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
On September 23, 2015, OPM raised its estimate of people whose fingerprints were stolen to 5.6 million and offered study and future notice. NIST’s 2017 and current identity guidance treat biometrics as non-secret and require a physical authenticator.
- The week On September 23, 2015, the Office of Personnel Management issued a statement on the theft of background investigation records.
- The inventory: how a count of 1.1 million became 5.6 million The September statement did not say an attacker had taken more data.
- The remedy: what services can and cannot repair The services OPM described in July were identity restoration, identity theft insurance, monitoring for minor children, continuous credit monitoring and fraud monitoring beyond credit files. [1] Each addresses misuse that shows up in financial or identity records, where it can be detected and repaired.
- The authenticator: a stolen image should not be a stolen key Fingerprints serve vetting, and in federal identity systems they also authenticate.
- The custodian: holding biometric data before anything is stolen FIPS 201-3 shows two different uses of the same prints.
- How the episode reaches banks and health systems NIST’s digital identity guidelines are written for federal agencies’ online services.
- The objections, taken in turn The September statement said federal experts believed the ability to misuse the data was limited. [2] A reader could conclude the fingerprint loss mattered little.
- Checks an agency CIO or privacy officer could run These checks are our suggestions.
- With hindsight The sources we opened do not report what the fingerprint working group concluded, so we do not state an outcome.
The week
On September 23, 2015, the Office of Personnel Management issued a statement on the theft of background investigation records. It said the number of people whose fingerprints had been stolen had risen from approximately 1.1 million to approximately 5.6 million. [2]
The statement explained how the number changed. OPM and the Department of Defense had been analyzing the affected data to verify its quality and completeness as they prepared to notify people. [2] In that process they identified archived records containing additional fingerprint data that had not previously been analyzed. [2]
The statement said the change did not increase the overall estimate of 21.5 million individuals affected by the incident. The fingerprints belonged to a subset of the people whose Social Security Numbers and other sensitive information had been taken. [2]
The earlier figures came from OPM’s announcement of July 9, 2015. That release said an interagency team had concluded with high confidence that sensitive information, including Social Security Numbers, of 21.5 million individuals was stolen from the background investigation databases. [1]
It broke that figure into 19.7 million people who had applied for a background investigation and 1.8 million non-applicants, predominantly spouses or co-habitants of applicants. [1] It said some records included findings from interviews by background investigators, and that approximately 1.1 million included fingerprints. [1]
The July release listed what the records contained: identification details such as Social Security Numbers, residency and educational history, employment history, information about family and acquaintances, and health, criminal and financial history. [1] It said usernames and passwords that applicants used to fill out their forms were also stolen. [1]
It said this incident was separate from but related to an earlier one, discovered in April 2015, in which personnel data for 4.2 million current and former federal employees was stolen. [1] It said anyone who underwent a background investigation through OPM in 2000 or afterwards was highly likely to be affected. [1]
The July release also described the help on offer. OPM and the Department of Defense would work with a private firm to provide identity restoration support, identity theft insurance, identity monitoring for minor children, continuous credit monitoring and fraud monitoring beyond credit files. [1] It said the services would be provided “for a period of at least 3 years, at no charge.” [1]
The September statement addressed fingerprints separately. It said federal experts believed that, as of then, the ability to misuse fingerprint data was limited, but that this probability could change over time as technology evolves. [2]
It said an interagency working group would review the potential ways adversaries could misuse fingerprint data now and in the future. [2] Its members were to include the FBI, the Department of Homeland Security, the Department of Defense and other members of the Intelligence Community. [2] It said the group would also seek ways to prevent such misuse. [2]
The statement made a forward promise. If new means of misusing the fingerprint data were developed in the future, the government would provide additional information to individuals whose fingerprints may have been stolen. [2]
It repeated that all affected individuals and their minor dependent children, as of July 1, 2015, were eligible for identity theft and fraud protection services at no cost. [2] It said OPM, with the Department of Defense, was working to begin mailing notifications on a rolling basis, and that an interagency team would keep refining the data. [2]
The OPM breach is remembered for its headline count of 21.5 million people. The September 23 revision told a narrower story. About 5.6 million of them, by the agency’s estimate, had lost something that a credit freeze, a new password or a new card number cannot replace. [1] [2]
Our argument is that the fingerprint count exposed a mismatch between the harm and the available remedies. The services OPM offered were built for misuse of identity data that can be monitored and repaired. The statement itself said the fingerprint risk was uncertain and could grow, and its answer was a working group and a promise of future notice. [1] [2]
NIST’s digital identity guidance later stated the underlying principle: biometric characteristics do not constitute secrets. [5] [6] In our reading, that principle is the right lens on what the theft changed and what it did not. Fingerprints that are not secrets cannot be stolen in the way a password is. They can still be misused, and they cannot be reissued.
We take the episode in four parts, each a different question an organization holding biometric data has to answer:
- The inventory: knowing which records contain biometric data, which is how OPM’s count changed.
- The remedy: what notification and identity services can and cannot do for a biometric loss.
- The authenticator: how systems that use fingerprints should be built so that a stolen image is not a stolen key.
- The custodian: what holding biometric data obliges an organization to do before anything is stolen.
The parts do not overlap. The first is knowledge, the second response, the third design and the fourth stewardship. After them we look at how the episode reaches banks and health systems, take the objections in turn and end with checks an agency CIO or privacy officer could run.
The inventory: how a count of 1.1 million became 5.6 million
Archived records not previously analyzed
The September statement did not say an attacker had taken more data. It said OPM and the Department of Defense had found, in records already stolen, additional fingerprint data that had not been analyzed before. [2] The change came from the agencies’ understanding of their own holdings.
The data in question was archived. In our reading, that word matters. Archived records sit outside the systems people use every day, and their contents can drift from what current inventories describe. An agency can know what its live database holds and still be unsure what its archives hold. [2]
The analysis that found the extra fingerprints was done to verify data quality and completeness before mailing notices. [2] That is a sound reason to look. It is also a late one. The inventory question was answered after the theft, under the pressure of notification.
Why biometric fields are hard to count
Fingerprints in an investigation file may not be stored as a labeled field. They can arrive as images, as cards scanned to files, or as templates produced for matching. Each form can sit in a different repository or archive. The July release described the records as containing fingerprints without saying in what form. [1]
For an inventory, that variety means a simple query will not find them all. A field called “fingerprint” in one schema is a scanned attachment in another and a template in a third. In our reading, counting biometric records requires knowing the formats as well as the tables. [2]
The same problem appears wherever biometric data enters through more than one path. A program that collects prints on paper, on live-scan devices and through a vendor will hold them in different shapes. Each path needs its own entry in the inventory.
Counts drive notice
The count is not an academic figure. It decides who receives which notice. The September statement said an interagency team would keep analyzing and refining the data as it prepared to mail notification letters. [2]
An agency that cannot say which individuals’ records contained fingerprints cannot tailor the notice to them. In OPM’s case the revision came while the letters were still being prepared. [2] A count that changes after notices go out would mean some people learn of a loss late, and others must be contacted twice.
In our reading, the point for data owners is that the inventory of biometric holdings is part of incident readiness. It is cheaper to build it before an intrusion than to reconstruct it from stolen copies afterwards.
Two incidents, two populations
OPM was handling two inventories at once. The July release described an earlier incident, discovered in April 2015, in which personnel data for 4.2 million current and former federal employees was stolen. [1] The background investigation incident was separate, and it reached applicants, spouses and co-habitants as well as employees. [1]
The two populations overlapped but were not the same. A person could be in one, the other or both, and the data lost differed between them. The July release said there was no evidence that separate systems holding health, financial, payroll and retirement records had been affected. [1]
In our reading, the episode shows why an inventory needs to record which population each system describes, as well as which fields it holds. A notice program built on the wrong population mails the wrong letter to the right person, or the right letter to the wrong one.
Resettable secrets in the same files
The July release said usernames and passwords that applicants used to fill out their background investigation forms were also stolen. [1] Those are the opposite of fingerprints: secrets that can be changed.
Their presence in the same breach makes the contrast concrete. For the passwords, the remedy is a reset and a check for reuse elsewhere. For the fingerprints, there is no reset. The same intrusion produced one loss that could be closed quickly and one that could not. [1] [2]
In our reading, an inventory that records whether each data element can be changed after a loss is a useful addition to the usual sensitivity labels. It tells responders, in advance, which parts of a breach can be repaired and which can only be managed.
Data quality as a security control
The statement framed the work as verifying the data’s quality and completeness. [2] Data quality can look like a reporting or analytics concern. In this episode it decided the accuracy of the government’s account to millions of people.
In our reading, that makes data quality a security control in the plain sense: it determines how well an organization can describe and contain a loss. Metadata on what each record contains, when it was collected and where copies live is what turns a breach into a bounded event.
The remedy: what services can and cannot repair
Services built for credentials that can change
The services OPM described in July were identity restoration, identity theft insurance, monitoring for minor children, continuous credit monitoring and fraud monitoring beyond credit files. [1] Each addresses misuse that shows up in financial or identity records, where it can be detected and repaired.
A stolen Social Security Number can be watched for in credit applications. A stolen password can be changed. A compromised account can be closed. The services fit those cases. [1]
A fingerprint fits none of them. It cannot be changed, and its misuse would not necessarily appear in a credit file. In our reading, the services were sound for the Social Security Numbers that every affected person lost. They were not designed for the biometric subset.
A promise of future notice
The September statement recognized the gap. It did not offer a service for fingerprints. It offered review, by an interagency working group, of how adversaries could misuse fingerprint data now and in the future, and a search for ways to prevent it. [2]
It also promised that if new means of misuse were developed, the government would provide additional information to the people affected. [2] That is an open-ended commitment. It has no end date in the statement, and it depends on keeping a reliable way to reach millions of people over time.
In our reading, a promise of future notice turns a one-time breach response into a long-lived program. Someone must own the list of affected people, keep contact details current and decide what development would trigger a further notice. The statement did not say who.
The promise also implies monitoring. To know when new means of misuse have been developed, someone must watch the research and the threat reporting on biometric spoofing and matching. The statement placed that work with the interagency group. [2]
Uncertainty stated honestly
The statement’s assessment of the risk was cautious in both directions. It said federal experts believed the ability to misuse the fingerprint data was limited as of then, and that this probability could change over time as technology evolves. [2]
That framing is useful because it does not claim more than was known. It also leaves the affected person with no action to take. A credit freeze is something a person can do. A changing probability of biometric misuse is not. [2]
In our reading, the honest statement of uncertainty is the right model for any organization facing a biometric loss. It should be paired with a clear owner for the follow-up, which is the part the statement left general.
What an affected person could do
For the Social Security Number loss, affected people had steps to take: enroll in the services, watch their credit and respond to alerts. The July release described those services and said OPM would send educational materials on preventing identity theft. [1]
For the fingerprint loss, the statement offered no step for the individual. It described what the government would do: study misuse, look for prevention and provide information if new means of misuse emerged. [2]
In our reading, that asymmetry is the clearest sign that biometric loss is a custodian’s problem more than an individual’s. The person cannot change the data or watch for its misuse. Only the systems that accept fingerprints can be changed, which puts the burden on their owners.
The duty to minors
The September statement extended eligibility for services to minor dependent children of affected individuals, as of July 1, 2015. [2] The July release had listed identity monitoring for minor children among the services. [1]
Children’s identity data has a long exposure window. The investigation files included information about family members, and the July release said the non-applicant population was predominantly spouses or co-habitants. [1] In our reading, the episode shows how vetting data reaches people who never applied for anything, and how the remedy has to follow them.
Notice by mail, at scale
The September statement said OPM and the Department of Defense were working to begin mailing notifications on a rolling basis. [2] Mail was the channel. The statement did not say where the mailing addresses came from. The July release said people investigated in 2000 or afterwards were highly likely to be affected, so some of the records were old. [1]
An address collected for an old investigation may not reach the person now. In our reading, that is the operational limit of any long-lived notice duty: the contact data ages with the rest of the record.
For any organization that holds biometric data, the ability to reach the subject is part of custody. Data that cannot be tied to a current way of reaching the person is harder to protect after a loss, because the person cannot be told. [2]
The authenticator: a stolen image should not be a stolen key
Fingerprints as authenticators in federal systems
Fingerprints serve vetting, and in federal identity systems they also authenticate. FIPS 201-3, the standard for Personal Identity Verification of federal employees and contractors, says the full set of fingerprints is collected for identification against FBI databases. [4]
It says the two mandatory fingerprints are used to prepare biometric templates stored on the PIV Card. Those fingerprints provide an interoperable authentication mechanism through off-card comparison, which the standard calls BIO or BIO-A. [4] It says they are also the primary means of authentication during PIV issuance and maintenance. [4]
The standard also describes optional on-card biometric comparison, OCC-AUTH, which it says was introduced in FIPS 201-2. [4] With on-card comparison, the template stays on the card and the match happens there.
Not secrets, by design
NIST’s authentication guideline, SP 800-63B, published in June 2017 (we read the copy that includes updates as of March 2, 2020), explained why it supports only limited use of biometrics. It said biometric characteristics do not constitute secrets. [5] It said they can be obtained online, by photographing someone, lifted from objects a person touches or captured with high-resolution images. [5]
From that premise it drew a rule. Biometrics “SHALL be used only as part of multi-factor authentication with a physical authenticator.” [5] A fingerprint, on this design, unlocks something the person holds. It does not stand alone.
In our reading, this design is what limits the damage of a fingerprint theft for authentication. If a system accepts a fingerprint only together with a physical authenticator, a stolen fingerprint image is not enough to log in. The OPM loss is serious for other reasons, but it is not a stolen master key to every system that uses fingerprints. [5]
Presentation attacks and matching thresholds
The 2017 guideline set conditions on the biometric component. It required a false match rate of 1 in 1000 or better under defined test conditions, and said the system should implement presentation attack detection, such as liveness detection. [5]
It said presentation attack detection was being considered as a mandatory requirement in future editions. [5] That note points at the exact risk the OPM statement left open. A stolen fingerprint image matters most if someone can build a fake that a sensor accepts.
In our reading, the controls that answer the September statement’s uncertainty are sensor-side: presentation attack detection, authenticated sensors and limits on failed attempts. The 2017 guideline required an authenticated protected channel between sensor and verifier, and limits on consecutive failed attempts. [5]
Revocation is limited
The 2017 guideline also addressed revocation. It said biometric template protection schemes provide a method for revoking biometric credentials comparable to other authentication factors, but that their availability was limited and standards for testing them were under development. [5]
That is the technical form of the point the OPM statement implied. A password can be reset. A biometric credential can be revoked only if the system stores it in a form designed for revocation. [5] Without that, the only revocation available is to stop trusting the biometric for that person.
For agencies, in our reading, the PIV design softens the problem. A card can be revoked and reissued even if the fingerprints on it cannot. The card, not the finger, carries the revocable credential. [4]
Issuance as the exposed step
FIPS 201-3 says the two mandatory fingerprints are also the primary means of authentication during PIV issuance and maintenance. [4] In our reading, that makes issuance and card maintenance the moments when the fingerprint carries the most weight, because the new card is not yet in hand.
Those are the steps where a stolen fingerprint image matters most for federal credentials. A forged finger that fooled an enrollment station could, in principle, help someone obtain a card in another person’s name. The standard puts several defenses at those steps. [4] It requires the applicant to appear in person at least once before issuance, at the issuing facility or a supervised remote identity proofing station. [4] It requires separation of duties, so that no single individual can issue a PIV Card without another authorized person. [4] Where identity proofing and enrollment take more than one visit, it requires an automated biometric comparison at each visit against data collected at a previous visit. [4]
For agency identity teams, this suggests where to look first. Day-to-day logins pair the fingerprint with the card. Issuance and maintenance stations are where sensor quality, presentation attack detection and attended operation carry the most weight.
Images, templates and where they match
FIPS 201-3 distinguishes the fingerprints collected for identification from the templates prepared for the card. [4] It also describes on-card comparison, in which the match happens on the card itself. [4]
Those distinctions are design choices with security consequences. A system that stores full images centrally holds the richest copy. A system that stores templates holds a derived copy. A system that matches on a card or device keeps the reference with the person.
In our reading, each step away from central images reduces what a single intrusion can take. NIST’s 2017 guideline described template protection schemes as a way to make biometric credentials revocable, while noting that their availability was limited. [5] The choice of where matching happens is the custodian’s lever when revocation is not available.
The custodian: holding biometric data before anything is stolen
Collection is not the same as storage
FIPS 201-3 shows two different uses of the same prints. The full set is collected for identification against FBI databases; two prints are used for templates on the card. [4] Each use needs a different storage decision.
An agency that collects ten prints for a background check does not need to keep ten-print images in every system that later touches the person’s record. In our reading, the question a custodian should ask is which systems need the raw images, which need templates and which need neither.
The OPM episode shows the cost of not asking. Fingerprint data sat in archived records of the background investigation systems, where it was available to be stolen with everything else. [2]
Encryption and system boundaries
GAO’s August 2017 review of OPM found that the agency had identified its high value assets, including systems with sensitive information that might attract adversaries. [8] It found that OPM did not encrypt stored data on one selected system and did not encrypt transmitted data on another. [8]
GAO said that until OPM completed government-wide requirements, its systems were at greater risk than they needed to be. [8] For biometric data, encryption at rest and in transit is the baseline control that keeps a copy of a database from being a copy of the prints.
In our reading, the high value asset designation is the right place to record biometric holdings. A system holding fingerprint images is attractive to an adversary by any measure. Naming it as such ties it to the stronger controls that designation brings. [8]
Retention after the purpose is met
Fingerprints collected for a background investigation serve a defined purpose: identification against FBI records and, for PIV, templates on a card. [4] Once the check is done and the card is issued, the question is how long the raw images need to remain in the investigation file.
The sources we opened do not set out OPM’s retention rules for those images, and we do not state them. The episode still frames the question. The September statement said the extra fingerprints were found in archived records. [2] Data that is archived rather than deleted remains exposed to whoever reaches the archive.
In our reading, a retention rule tied to the purpose of collection is the cheapest control a custodian has. It reduces what can be stolen without depending on any detection or encryption working.
Contractors who run the systems
GAO also looked at contractor-operated systems. It found OPM’s procedures did not ensure that security controls were comprehensively tested in assessments of those systems. [8] It said OPM’s policy on quality assurance reviews did not include detailed guidance on how the reviews were to be done. [8]
Biometric collection and storage can involve contractors at several points: enrollment devices, card production, hosting and investigations support. Each can be a place where the data is copied. In our reading, the custodian’s duty reaches through those contracts, and assessments of contractor systems must test the controls rather than list them.
GAO said OPM concurred with four of its five public recommendations and partially concurred with the one on validating corrective actions. [8] That one, validating that fixes work, is the step that turns a remediation plan into evidence.
Sensitive personal information by rule
The current revision of NIST’s authentication guideline, SP 800-63B-4, which says it supersedes SP 800-63B, makes the custodial duty explicit. It says biometric data “SHALL be treated and secured as sensitive personal information.” [6]
It notes that the use of biometric characteristics, especially when stored for central verification, introduces new privacy concerns. [6] Central storage is what OPM had: a repository of prints that could be taken in bulk.
In our reading, the current guideline points custodians toward designs that avoid central stores where they can, such as comparison on a card or device. [4] [6] Where they cannot, it points toward treating any central store as a high-value target. [4] [6]
How the episode reaches banks and health systems
Banks and credit unions
NIST’s digital identity guidelines are written for federal agencies’ online services. The current revision says it focuses on authentication of subjects who interact with government information systems over networks. [6] The 2017 edition says it may be used by nongovernmental organizations on a voluntary basis. [5] In our reading, it does not bind banks or credit unions.
Financial institutions that use biometrics can still learn from its reasoning. The current revision says biometric comparison based on voice SHALL NOT be used. [6] A bank that authenticates callers by voice is making a different choice from the one NIST makes for federal systems. In our reading, that choice should be a deliberate one, recorded with its compensating controls.
The same revision requires that an alternative non-biometric authentication option always be provided to the subscriber. [6] For a bank, an alternative path is also a fallback when a customer’s biometric is compromised or no longer works.
Health systems
Health systems may use biometrics to identify patients or to control staff access to systems and medication. The OPM episode is relevant to them as custodians: what they store, in what form, and whether a stolen copy would be usable.
The principle from the federal guidance carries over in our reading. A biometric used only to unlock a credential the person holds is a lower-risk design than a central database of images used to identify people at any terminal. [5] [6] Health systems should check with counsel on state laws that may govern biometric data, which this article does not address.
State and local agencies
State and local agencies collect fingerprints for licensing, employment and benefits. They hold the same kind of data OPM lost. The FIPS 201-3 split between full collection for identification and templates for authentication is a useful model for their own storage decisions. [4]
In our reading, the OPM episode is a reason for those agencies to inventory where fingerprint images are kept after a check is complete, and to decide whether they need to keep them at all.
Vendors that enroll and match
In every sector, biometric systems involve vendors: sensor makers, enrollment services, matching software and hosted identity platforms. Each may hold a copy of the data or the templates derived from it.
The GAO findings on OPM’s oversight of contractor-operated systems apply by analogy. GAO found OPM’s assessments did not ensure comprehensive testing of controls on those systems. [8] In our reading, any organization that outsources biometric enrollment or matching needs contract terms on where data is stored, whether images are kept and how deletion is proven.
Those terms are also where an organization learns whether its design matches its intent. A program meant to keep templates on cards can still produce central image stores at a vendor if the contract is silent.
The objections, taken in turn
“The fingerprints were low risk”
The September statement said federal experts believed the ability to misuse the data was limited. [2] A reader could conclude the fingerprint loss mattered little.
The same sentence said the probability could change over time as technology evolves. [2] The risk was not described as low and stable. It was described as limited for now and uncertain later. In our reading, that is a reason for design controls that do not depend on the risk staying low.
“Authentication was never at stake”
A second objection is that federal systems did not rely on fingerprints alone, so the theft could not be used to log in. [5] Under the 2017 guideline’s rule that biometrics be used only with a physical authenticator, that is correct for systems following it. [5]
But the guideline is a design standard, and not every system using fingerprints follows it. The 2017 text also said presentation attack detection was not yet mandatory. [5] In our reading, the objection holds for well-built systems and should prompt a check of the rest.
“Monitoring services were the best available”
A third objection is that OPM did what it could. There was no service that could protect a fingerprint, and the government offered what existed. [1] [2]
We agree with the premise. The point is about expectations. The services addressed the Social Security Number loss, which every affected person shared. For the fingerprint subset, the honest answer was the one the statement gave: study, prevention research and a promise of future notice. [2] Organizations holding biometrics should know in advance that this is the shape of the answer they would have to give.
“Biometrics are safer than passwords”
A fourth objection is that biometrics resist phishing and reuse in ways passwords do not, so the lesson should not discourage them.
We do not argue against biometrics. The NIST guidelines support their use within limits, paired with a physical authenticator and with presentation attack detection. [5] [6] The argument is narrower: use them so that a stolen copy does not matter, and hold them so that a stolen copy is less likely.
“Only a capable adversary could use them”
A fifth objection is that misuse of stolen fingerprints would need resources few attackers have, so the risk to ordinary systems is small. The statement’s choice of working group members, including the Intelligence Community, points to the kind of adversary its authors had in mind. [2]
That may be so, and the statement did not name the attacker, nor do we. [2] But the statement also said capability could change as technology evolves. In our reading, a risk that depends on an adversary’s tools is a risk whose size is set by the slowest-moving part of the defense, which for biometrics is the data that cannot be changed.
The practical answer is the same as for the other objections. Design the systems that use fingerprints so that an image alone is not enough, and keep fewer images.
Checks an agency CIO or privacy officer could run
These checks are our suggestions. They are not requirements of the OPM statement, FIPS 201-3 or NIST’s guidelines, though some draw on them. Agencies and other organizations should confirm their obligations with counsel and their own policies. [4] [6]
- Biometric inventory: every system and archive holding fingerprint images, templates or other biometric data is listed, with format and owner.
- Archive review: archived records have been checked for biometric content, not assumed from the live schema.
- Purpose test: each store of raw images has a stated need, and stores without one have a deletion plan.
- Pairing rule: systems that accept a biometric require a physical authenticator with it.
- Sensor defenses: biometric systems record whether presentation attack detection is in place, and where it is not.
- Alternative path: every biometric login has a non-biometric option.
- Encryption: biometric stores are encrypted at rest and in transit, and named as high value assets where that designation applies.
- Contractor testing: assessments of contractor systems that hold biometrics test controls as well as describe them.
- Contact data: the organization can reach the people whose biometric data it holds, years after collection.
- Retention rule: raw biometric images have a written retention period tied to the purpose that justified collecting them.
- Follow-up owner: if a biometric loss occurred, someone owns the list of affected people and the decision on further notice.
In our reading, OPM’s September 2015 statement showed an agency learning the extent of what it had held only after it was gone. Its remedy could not reach the most permanent part of the loss. [2] The checks above are about knowing what is held, holding less of it and designing systems so that a copy is not a key.
With hindsight
The sources we opened do not report what the fingerprint working group concluded, so we do not state an outcome. What they do record is how the remedy for the wider loss was later set in law, and how the law treated the fingerprints.
Congress set the terms of continued services in the Consolidated Appropriations Act, 2017, approved on May 5, 2017. Section 633 of its Financial Services and General Government division required OPM to provide affected individuals with complimentary identity protection coverage. [7]
As enacted, the section said the coverage must be no less comprehensive than what OPM had provided before enactment. [7] It had to be effective for a period of not less than 10 years, and include not less than $5,000,000 in identity theft insurance. [7] It defined affected individuals as those whose Social Security Numbers were compromised in the breach of personnel records that OPM announced on June 4, 2015, or in the breach of OPM systems holding background investigation information. [7]
The section opened with the words “For fiscal years 2016 through 2026,” and then set those terms. [7] Its definition turned on Social Security Numbers, not fingerprints. In our reading, the law extended the remedy built for identity data and did not create one for biometric data. [7]
Congress later amended the section. Section 630 of division E of the Consolidated Appropriations Act, 2018, approved on March 23, 2018, struck the paragraph that required coverage effective for a period of not less than 10 years, and renumbered the insurance requirement as the second paragraph. [3] It did not change the opening words that set the fiscal years. [3] As checked on 2026-11-16, the section as amended in 2018 still ends its term with fiscal year 2026, and we did not find an official text of a later law that changes it. [3] [7]
GAO reported on OPM’s security on August 3, 2017. It found OPM had completed actions on 11 of 19 recommendations US-CERT made after the breaches and had taken actions on the other 8, with 4 of those needing further improvement. [8] It found OPM did not validate corrective actions to ensure they addressed the recommendations. [8]
NIST published SP 800-63B in June 2017. It said biometric characteristics do not constitute secrets and required that biometrics be used only with a physical authenticator. [5] Its current revision, SP 800-63B-4, as checked on 2026-11-16, keeps that rule, requires a non-biometric alternative, bars voice-based biometric comparison and requires biometric data to be treated as sensitive personal information. [6]
The current revision also tightened the matching requirement. It requires a false match rate of one in 10000 or better for all demographic groups, and requires presentation attack detection for facial recognition. [6] The 2017 text had set 1 in 1000 and listed presentation attack detection as a recommendation. [5]
FIPS 201-3, issued in January 2022, keeps fingerprints at the center of federal PIV credentials, as templates on the card and as the primary means of authentication during issuance. [4] In our reading, the card design means the revocable credential is the card, not the finger. [4]
What held up from September 2015 is the statement’s caution: the risk was uncertain and could grow, and the government’s answer was study and future notice. [2] What changed is the guidance around biometrics, which now treats them as non-secret, demands a physical authenticator and treats stored biometric data as sensitive. [5] [6]
In our view, the fingerprint count is best remembered as the moment a large federal data holder found it could not repair a loss, only contain its consequences. The design rules that followed aim to make the next such loss matter less.
Spotted an error? Tell us and we will check it and correct the article where needed.