Equifax’s September 2017 disclosure put the answers to agencies’ online identity questions in doubt

The consumer reporting agency said on September 7, 2017 that attackers had accessed names, Social Security numbers, birth dates and addresses. In our reading, the federal exposure ran through online sign-up checks that quizzed applicants on credit-file history.

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

The consumer reporting agency said on September 7, 2017 that attackers had accessed names, Social Security numbers, birth dates and addresses. In our reading, the federal exposure ran through online sign-up checks that quizzed applicants on credit-file history.

  1. The week On Thursday, September 7, 2017, the consumer reporting agency Equifax Inc. announced a cybersecurity incident that it said potentially affected approximately 143 million U.S. consumers.
  2. The question: how credit-file questions became an identity gate GAO’s 2019 report split remote identity proofing into two major steps.
  3. The contract: what an agency was owed when its identity vendor was breached GAO reported that Equifax did not directly notify its major federal customers before the public announcement.
  4. The replacement: what moving off the questions cost and whom it left behind GAO’s 2018 report described the Postal Service’s change.
  5. The standard: NIST’s 2017 limits and why guidance alone did not end the practice NIST’s June 2017 guideline did not ban KBV outright.
  6. How the exposure reached different organizations IRS used identity checks to protect Get Transcript, which GAO said lets users view tax returns and account transcripts and download wage and income documents.
  7. Objections, in order The release did say that.
  8. What an identity program owner could check The list below is a set of checks we would run today.
  9. With hindsight GAO published its account of the breach and the federal response in August 2018.

The week

On Thursday, September 7, 2017, the consumer reporting agency Equifax Inc. announced a cybersecurity incident that it said potentially affected approximately 143 million U.S. consumers. The announcement was filed with the Securities and Exchange Commission as an exhibit to a current report on Form 8-K. It said criminals had exploited a U.S. website application vulnerability to gain access to certain files.

The release set out a timeline. Based on the company’s investigation, the unauthorized access occurred from mid-May through July 2017. The company said it discovered the access on July 29 and acted at once to stop the intrusion. It said it had engaged an independent cybersecurity firm for a forensic review and had reported the criminal access to law enforcement.

The release then listed what had been reached. The information accessed primarily included names, Social Security numbers, birth dates, addresses and, in some instances, driver’s license numbers. Credit card numbers for approximately 209,000 U.S. consumers were accessed. So were certain dispute documents with personal identifying information for approximately 182,000 U.S. consumers. The company also found unauthorized access to limited personal information for certain UK and Canadian residents.

One line concerned the company’s main business. The company said it had found no evidence of unauthorized activity on its core consumer or commercial credit reporting databases. It said its investigation was substantially complete but ongoing, and expected to finish in the coming weeks.

The response the release described was aimed at consumers. The company set up a dedicated website where people could check whether their information had potentially been affected. The same site let them enroll in credit file monitoring and identity theft protection, free for one year to all U.S. consumers.

The package covered monitoring of all three national credit reports, copies of the company’s own credit reports, and the ability to lock and unlock those reports. It added identity theft insurance and internet scanning for Social Security numbers. A dedicated call center would be open every day, including weekends.

Two further channels were promised. The company would send direct mail notices to consumers whose card numbers or dispute documents had been affected. It said it was contacting U.S. state and federal regulators and had sent written notifications to all U.S. state attorneys general.

The release did not mention federal agencies as customers. Yet the fields it listed were the starting point for how agencies proved identity online. In June 2017, the National Institute of Standards and Technology had published Special Publication 800-63A, its guideline on enrollment and identity proofing for federal systems.

That guideline described a step called knowledge-based verification, or KBV. An applicant answers questions that only the real person should be able to answer. NIST set limits on it. At the identity assurance level it called IAL2, a credential service provider could not use KBV to verify an identity against more than one piece of validated evidence.

The guideline also restricted where the questions came from. A provider was to use only information expected to be known only to the applicant and the authoritative source. That included any information needed to begin the KBV process. Information accessible freely, for a fee in the public domain, or via the black market was not to be used.

Where questions were used, NIST required at least four, each needing a correct answer. It recommended free-form answers, set a two-minute inactivity timeout per question and capped attempts at three. It barred questions whose answers never change. It recommended checking knowledge of recent transactions in which the provider itself took part. Applicants had to be able to opt out of KBV and use another process.

So by the week of the announcement, federal guidance had already narrowed the role of quiz-style identity checks. The September 7 release listed the very fields that open such a check. It said nothing about what an impostor could do with them at an agency’s front door. That question fell to the agencies that bought identity verification services from consumer reporting agencies, and that still verified applicants by quizzing them.

We read the September 7 announcement first as a consumer protection story, and with reason. The release offered monitoring to all U.S. consumers, whether or not their own data had been reached. People could not change their birth dates or Social Security numbers once those had been exposed.

The same fields mattered in a second place: the online front doors of federal programs. Several agencies verified people remotely by asking questions drawn from credit files. A consumer reporting agency generated the questions from its records. The applicant answered. The vendor scored the answers, and the agency treated a passing score as proof of identity.

Our argument is that the breach exposed how much of that front door rested on information the public did not control. The agencies’ response turned on four matters they had to settle one by one. We take them as four parts:

  1. The question: how credit-file questions became an identity gate.
  2. The contract: what an agency was owed when its identity vendor was breached.
  3. The replacement: what moving off the questions cost and whom it left behind.
  4. The standard: NIST’s 2017 limits and why guidance alone did not end the practice.

The first part is the mechanism. The second is the commercial relationship that framed the response. The third is the operational change that followed. The fourth is the governance that was supposed to drive it. After the four parts we look at how the exposure differed across agencies and sectors, at objections, and at what an identity program owner could check today.

A note on sources and our reading. We describe the week from Equifax’s own release and from NIST’s June 2017 guideline. We describe agency responses from two later reports by the Government Accountability Office, from August 2018 and May 2019. Statements about what guidance required are our reading. Agencies and firms should rely on current NIST and OMB text and their own counsel.

The question: how credit-file questions became an identity gate

Resolution and verification are different steps

GAO’s 2019 report split remote identity proofing into two major steps. Resolution decides which identity a person is claiming. Validation and verification check that the person is who they claim to be. An agency resolves an identity by comparing a name, address, date of birth and Social Security number with records it holds or with a vendor’s records.

Resolution on its own proves little. Anyone holding a person’s name, birth date and Social Security number can complete it. Verification is the step meant to stop that impostor. In the agencies GAO reviewed, knowledge-based verification was the method used for that second step, at least for some services.

The two steps can look like one screen to an applicant. The person types identifying details, then answers a few questions, then gets an account. Behind the screen, the first half matches a record and the second half tests knowledge. The breach struck the first half directly.

Where the questions came from

GAO described how this worked at the Social Security Administration. Applicants to the my Social Security service gave identifying information and answered questions generated by a consumer reporting agency. SSA’s system relayed the answers to the vendor’s identity proofing system. That system compared them with its records and sent back a result.

SSA decided how many questions to ask and how many right answers it required. The Postal Service ran a similar flow for its Informed Delivery service, with questions from a consumer reporting agency or another provider. CMS used the method before granting access to Healthcare.gov accounts.

The Department of Veterans Affairs used a Defense Department logon service that relied on a consumer reporting agency to generate questions for some applicants. In each case, the questions drew on personal and financial history held by a private company. The vendor, not the agency, scored the answers against its own records.

Why the breach mattered even without the core databases

Equifax said it had found no evidence of unauthorized activity on its core credit reporting databases. One could read that as good news for identity checks, since the questions draw on credit histories. In our reading, the reassurance answered a narrower question than the one agencies faced.

A KBV session starts with resolution. The impostor must first supply a name, date of birth, address and Social Security number that match. Those were the fields the release said had been accessed. Holding them moves an impostor past the first gate and into the question stage.

NIST’s guideline spoke to exactly this. It required that information needed to begin the KBV process be expected to be known only to the applicant and the authoritative source. After September 7, that condition was harder to claim for the identifying fields of a very large number of people.

The questions themselves draw on loans, addresses and accounts. Some of that history can be found in other places: earlier breaches, public records and data resellers. NIST had already barred questions built from information available for a fee or on the black market. The breach enlarged the pool of information that could end up in the second category.

GAO later put the point directly. In its May 2019 report, it said data stolen in breaches such as Equifax’s could be used fraudulently to answer knowledge-based verification questions. It said that risk led NIST to issue guidance in 2017 that effectively prohibits agencies from using KBV for sensitive applications.

Static answers age badly

A password can be changed after a breach. A birth date cannot. Resolution data is therefore exposed for good once stolen. The question stage was the only part of the check an agency could strengthen, and its answers were drawn from the same kind of personal history.

NIST had anticipated part of this. It barred questions whose answers do not change, such as a first car. It preferred questions about recent transactions with the provider itself, because only the applicant and the provider would know them. A credit-file question about an old lender or a past address falls outside that preference.

The distinction is between a secret the provider created and a fact the provider looked up. A one-time code sent by the agency, or a small deposit it made, is a secret it created. A past mortgage lender is a fact it looked up from someone else’s file. The first can be reissued after a breach. The second cannot.

What the agency could and could not see

GAO described the vendor’s system sending the agency a response on whether the applicant had answered correctly. In our reading, the signals that would show whether the check had weakened, such as how widely the underlying facts had leaked or repeated attempts against one identity, sat mainly with the vendor, if they were collected at all.

That split shaped the response. An agency wanting to know whether its check had weakened needed data from the vendor that had just been breached. GAO reported that the Internal Revenue Service and SSA obtained lists of affected individuals from Equifax. They matched the lists against their own customers and looked for possible identity fraud.

The lists did not say whether a given person had already been impersonated. They said whose data had been reached. An agency could then watch those accounts more closely, add steps for them, or warn them. Each choice put cost or friction on people who had done nothing wrong.

Fraud risk followed the data, not the contract

Information security officials at IRS, SSA, USPS and the Department of Homeland Security told GAO they were concerned about how the breached data could be used. Their concern included fraudulent access to government services from agencies that were not direct Equifax customers. An agency using a different vendor still faced the same stolen identifying fields.

In our reading, this is why the week mattered beyond Equifax’s customer list. The weakness sat in the method, not in one vendor’s systems. Any agency that opened a session with identifying data and closed it with history questions faced the same exposure, whoever generated the questions.

The applicant’s side

For an honest applicant, KBV is a test of memory about one’s own credit history. People with thin credit files may get fewer usable questions. People who fail are sent elsewhere: to a mailed letter, an office visit or a phone line. At CMS, GAO reported, applicants who could not complete the online process were directed to submit information by mail for manual review.

That fallback path is part of the identity system, in our reading. If a breach makes the online check weaker, an agency may raise its thresholds. More honest applicants then fail and move to the slower channels. Customer experience and fraud control pull on the same lever.

The same lever moves the other way. An agency under pressure to cut call volumes may lower its thresholds or accept more partial matches. That helps honest users and impostors alike. The decision belongs with a named owner who can see both the fraud data and the service data.

The contract: what an agency was owed when its identity vendor was breached

Notice came with the public announcement

GAO reported that Equifax did not directly notify its major federal customers before the public announcement. Officials from IRS and SSA told GAO why. Their contracts required notification only of breaches directly involving the systems that provided services to the federal government. The breached dispute portal was not such a system.

Those terms followed a familiar logic. They protected federal data held by the vendor. They did not treat the vendor’s other data as part of the federal service. Yet the identity service depended on that other data staying secret.

Scope written around systems, risk running through data

In our reading, this is the central contract point of the week. An identity verification service is a data product. Its value depends on the vendor’s whole data estate, not on the slice of systems that touch the government. A breach anywhere in that estate can weaken the service, even when no federal system is reached.

A notification clause scoped to systems that serve the government misses that dependency. It suits a hosting contract, where the risk is to the data the agency hands over. It does not suit a service whose output is a judgment about whether someone is who they claim to be. That judgment rests on information the agency never handled.

The same reasoning applies to any outside service that vouches for people: credit checks for loan applicants, background checks for staff, eligibility checks for benefits. Each rests on the provider’s data. A breach of that data weakens what the provider vouches for. Contract scope should follow what the service depends on.

What the agencies changed

GAO reported that IRS and SSA changed their contracts to require prompt notice of any future breach, among other things. SSA officials said the agency needed to know of any breach, whichever company system it affected. IRS made a similar change to its contract for credit reporting services.

The IRS change also shortened the clock. The contract now required notice within one hour after a breach is discovered, rather than within the previous 24 hours. IRS officials said cybersecurity language in the contract was modified to improve implementation and oversight of technical controls. After its assessment, IRS also updated its internal contractor requirements and controls for incident handling.

A one-hour clock is demanding. It assumes the vendor knows quickly that something is a breach and that a named contact can receive the notice at any hour. In our reading, the value of such a clause lies less in the hour than in the scope change beside it. A fast notice about the wrong systems is still the wrong notice.

Assessments on site

All three agencies GAO studied sent staff to Equifax’s data centers. SSA officials said their review assessed compliance with the NIST control baseline for systems rated as moderate impact. USPS reviewed physical and cybersecurity controls at the Alpharetta, Georgia, and St. Louis, Missouri, sites. IRS assessed the Alpharetta center and reviewed the St. Louis site separately.

Officials from all three told GAO their reviews did not uncover any major new problems. The reviews did identify lower-level technical concerns that the agencies required Equifax to address. An on-site review after a breach tests the controls as they stand. It cannot show how the breach happened, which was the subject of the company’s own forensic work.

Three agencies, three audits

The agencies acted independently. They told GAO it was unclear whether any single federal agency was responsible for coordinating the government’s response to a private-sector breach of this type. Each therefore planned and ran its own review.

In our reading, the separate reviews were rational for each agency and costly for the government as a whole. Three teams examined overlapping controls at the same vendor. Each produced its own findings. SSA shared its results with IRS, the Office of Management and Budget, the Social Security Subcommittee of the House Ways and Means Committee and the Senate Committee on Finance, which reduced some duplication after the fact.

A shared assessment model would have needed someone to own it. Nothing in GAO’s account shows such a model for identity data services bought under separate contracts. That gap was organizational, not technical, and it showed only when a shared vendor failed.

DHS offered help that was not taken

GAO reported that DHS, as the center for federal incident prevention and response, offered assistance to Equifax. Equifax reportedly declined because it had already retained an outside cybersecurity consultant. That left the agencies relying on their own reviews and on what the company chose to share.

The procurement in progress

IRS was already replacing its taxpayer identity verification contract when the breach was found. In June 2017 it had issued a request for quotations to three consumer reporting agencies holding federal supply schedule contracts. It selected another vendor’s quotation as the lowest-priced and technically acceptable. Equifax protested to GAO on July 5, 2017.

On September 29, while the protest was pending, IRS awarded Equifax a short-term, sole-source order for $7.25 million to keep the services running. GAO reported that IRS considered the services critical and unable to lapse. After its security assessments, IRS issued a stop-work order suspending performance under that order. GAO denied the protest on October 16, 2017, and IRS proceeded with the other vendor.

Continuity against confidence

That sequence shows the bind. An identity service that gates tax records cannot simply stop. Yet continuing with a breached vendor, even briefly, needed a reason. IRS used a short-term order to bridge the gap and an assessment to decide whether to stop.

In our reading, the contract lesson is to plan that bridge before it is needed. A second qualified source, or a manual path that can carry the load for a time, turns a forced choice into a managed one. Without either, an agency weighs a service outage against a known weakness under public scrutiny.

Who held the data about the people affected

The lists of affected individuals were Equifax’s records. IRS and SSA obtained them to protect their own customers. That required the vendor to share personal data with the agencies, for a fraud purpose the original contracts may not have contemplated.

Contracts for identity services can settle this in advance. They can say what breach data the agency may receive, how fast, in what format, and for what uses. They can say how long the agency may keep it. Settling those terms under pressure, after a breach, is slower and less careful.

The replacement: what moving off the questions cost and whom it left behind

USPS removed questions and mailed codes

GAO’s 2018 report described the Postal Service’s change. After its assessment, USPS consulted NIST about the risks of the knowledge-based questions it had been using with Equifax’s identity proofing service. It then removed certain questions. It added a procedure in which customers receive a code by mail to verify their mailing addresses.

A mailed code proves control of an address, not knowledge of a history. It is slower than an online quiz. For the Postal Service it drew on a system the agency already ran, which made it a natural choice there. For other agencies, a mailed code means paying for printing, postage and the delay before an account opens.

IRS and GSA used phones and mail

By the time of GAO’s 2019 review, IRS had changed its Get Transcript service. The applicant gave a phone number, which IRS checked through a consumer reporting agency against phone company records. IRS then sent a one-time PIN by text message. Applicants who could not pass that way received a confirmation PIN by postal mail.

The General Services Administration’s Login.gov asked for a photo of an identity document, such as a driver’s license, which GSA compared with information it obtained. It verified a phone number through a consumer reporting agency or another third party, then confirmed possession with a one-time code. Applicants who could not be verified that way received a PIN by mail.

At the time, Login.gov did not offer in-person proofing itself. GAO reported that GSA was working with USPS on a process for in-person proofing at post office locations. For individuals who could not be verified online, the agency whose service they wanted had to verify them.

Each alternative leaned on another dataset

In our reading, none of these replacements removed dependence on outside records. Phone verification relied on carrier records reached through a consumer reporting agency. Document checks relied on the issuing source’s data. Mailed codes relied on the address of record. What changed was the kind of evidence: possession of a phone or a mailbox, rather than knowledge of a history.

Possession evidence carries its own weaknesses. Phone numbers can be ported or taken over. Mail can be intercepted or redirected. Document images can be altered. NIST’s 2017 framework set requirements for those methods too. The point for an agency was that the risk profile changed, and its monitoring had to change with it.

Who failed the new checks

GAO reported the reasons officials gave for not adopting alternatives. They cited high costs and implementation challenges for certain segments of the public. Mobile device verification might not always be viable, GAO said, because not all applicants have mobile devices that can be used to verify their identities.

That is a customer experience fact with fraud consequences. An applicant without a smartphone, without a stable address or without a credit history meets friction under each method. Under KBV, a thin file means few usable questions. Under phone checks, a number may not match carrier records. Under mailed codes, a person who moves frequently may not receive the letter.

The applicants who fail are not spread evenly, in our reading. Benefit programs serve people for whom one of the three anchors may be missing: no smartphone, no fixed address, or little credit history. A check that suits taxpayers filing from a home computer may not suit them.

In-person proofing as the backstop

SSA allowed applicants who could not, or chose not to, complete the online process to prove their identity in person. USPS told GAO it planned to expand in-person proofing to most post offices and postal carriers in fiscal year 2019. GSA was working with USPS on in-person proofing for Login.gov.

In-person proofing is a strong fallback, and it costs staff time for every transaction. It also asks the applicant to travel. For benefits that serve elderly, disabled or rural people, travel can be a heavy requirement. The cost of a stronger online check is paid partly at the counter and partly by the applicant.

Measuring the trade-off

In our reading, an agency changing its check after the breach needed to measure two things at once: fraud blocked and honest applicants lost. The second is harder to see. A person who fails online and never calls leaves no record of an unmet need.

The minimum evidence is a set of rates by channel. The share that starts the online check, the share that passes, the share that fails at each step, and the share of failures that later finish through mail, phone or in person. Those numbers let an agency see whether a change shifted fraud or shifted people.

A second measure looks after enrollment. Accounts opened by impostors show themselves later: changed direct deposit details, address changes, requests for replacement documents. Tracking those events by enrollment method tells an agency which front door is letting impostors in.

Costs fell on different budgets

The identity service sits in an IT or security budget. The fallback paths sit in operations: call centers, field offices and mailrooms. A change that reduces fraud in one ledger may add staff time in another. GAO’s reference to costs reflected that split.

Agencies weighing alternatives had to cost the whole journey, not the vendor fee alone. That includes the people who give up. A benefit not claimed does not appear as a cost in any budget, but it is a cost to the program’s purpose.

The standard: NIST’s 2017 limits and why guidance alone did not end the practice

What the guideline required

NIST’s June 2017 guideline did not ban KBV outright. It limited KBV’s use in verification at IAL2 to a single piece of validated evidence. It said nothing restricted KBV for identity resolution. It forbade questions built from information available freely, for a fee in the public domain or on the black market. It required an opt-out, and set rules for the number of questions, timeouts and attempts.

Read against the September 7 release, one clause stands out. Information needed to begin the KBV process had to be expected to be known only to the applicant and the authoritative source. In our reading, a check that opened with identifying fields of the kind Equifax listed could no longer rely on that expectation for the people affected.

GAO’s reading in 2019

GAO said NIST’s 2017 guidance effectively prohibits KBV for sensitive applications. It then counted where six agencies stood. GSA and IRS had eliminated KBV for Login.gov and Get Transcript. VA had implemented alternatives for part of its process. SSA and USPS intended to reduce or eliminate KBV but had no specific plans. CMS had no plans to reduce or eliminate it.

Those six agencies had major public-facing web applications. GAO chose them as a non-probability selection, so the count describes them, not the whole government. Still, the spread is telling. By May 2019, the six agencies sat at four different points: two had eliminated KBV, one had partly replaced it, two intended to reduce it without plans, and one had no plans.

Why guidance did not drive change

GAO pointed to two gaps. NIST’s guidance did not tell agencies how to implement alternative methods with available technology for all segments of the public. It did not discuss the advantages and limits of current technologies or recommend which to adopt. Agencies told GAO they could not implement the guidance because of those technology limits.

The second gap was oversight. GAO said OMB had authority under the Federal Information Security Modernization Act to issue guidance. OMB had not issued guidance requiring agencies to report progress on NIST’s identity proofing guidance. Its planned identity management guidance, GAO said, did not require such reporting either.

In our reading, a technical standard without a reporting duty leaves each agency to weigh the standard against its own service pressures. An agency serving a broad public will weigh exclusion risk heavily. GAO reported that HHS, responding for CMS, disagreed with its recommendation. HHS did not believe the available alternatives were feasible for the individuals CMS serves.

Recommendations and responses

GAO recommended that CMS, SSA, USPS and VA develop plans to discontinue KBV. It recommended that NIST supplement its technical guidance with implementation guidance. It recommended that OMB require agencies to report their progress in adopting secure identity proofing practices.

Commerce, for NIST, and SSA, USPS and VA agreed with the recommendations. HHS disagreed. OMB provided a technical comment, which GAO incorporated, but did not comment on the recommendation. GAO said it continued to believe CMS should plan to discontinue KBV, since a variety of alternatives existed.

Standards set the floor, programs set the pace

In our reading, the gap between the 2017 standard and 2019 practice did not come from ignorance. Agencies knew the guideline. They faced a real choice between excluding some honest applicants and accepting a known fraud risk.

A standard can define the risk. It cannot by itself fund the counter staff, the mail operation or the in-person network that make an alternative humane. Those depend on budgets, partners and time. A program director weighing them is making a service decision as well as a security one.

The authoritative source question

NIST’s preference for transactions in which the provider took part points to a different design. An agency with its own records can ask about them. SSA knows a beneficiary’s payment history. IRS knows prior-year return figures. Those are not secret from all parties, but they are not on a credit file either.

Using agency records has its own limits. A first-time applicant has no history with the agency. Records can also leak. The design choice is between dependencies, not between risk and safety. What the week showed is that a dependency on a private credit file was a dependency on that company’s security as well.

How the exposure reached different organizations

Tax administration

IRS used identity checks to protect Get Transcript, which GAO said lets users view tax returns and account transcripts and download wage and income documents. That is information an impostor could, in our reading, put to use in tax fraud. Stolen identifying data raises the value of any weak step in that chain. By GAO’s 2019 review, IRS had moved Get Transcript to phone and mail verification.

IRS also bought credit reporting services from Equifax under a separate contract. That was the contract it amended for one-hour notice. A tax agency can hold several relationships with one vendor, each with different terms. A breach tests all of them at once.

Benefit agencies

SSA’s online service allowed people to request replacement Social Security or Medicare cards, check the status of applications, see statements and obtain benefit verification letters. An impostor with access touches income records and identity documents. SSA told GAO it intended to reduce or eliminate KBV but had no specific plan then.

SSA also issued a public blog post reminding people of steps to protect their Social Security numbers. Public guidance is a modest control. It reaches people who read it, not the ones an impostor has already chosen.

Health coverage

CMS used KBV before granting access to Healthcare.gov, where people apply for coverage and income-based subsidies. Failed applicants went to a mail-based manual review. HHS, for CMS, disagreed that alternatives were feasible for the population served. In our reading, that position put access for the people CMS serves ahead of the fraud risk GAO described.

Postal and address services

USPS used KBV for Informed Delivery, which shows images of incoming letter mail. For online change of address, it checked, through a third-party provider, that the address tied to the credit card used to pay the fee matched the old or new address. People without a card or who failed the online process could go to a post office.

Address control is itself a credential for other agencies, since mailed PINs depend on it. A weak check on change of address can undermine every agency that mails a code. In our reading, that made the Postal Service’s identity choices a shared dependency, whether or not other agencies counted it as one.

Veterans

VA relied on a Defense Department logon service that used KBV for some applicants. For others it used a commercial service combining KBV with document photos. GAO counted VA as having partly moved to alternatives. VA agreed to plan the rest.

State and local agencies

GAO’s reviews covered federal agencies only. States run unemployment insurance, Medicaid eligibility, motor vehicle and benefit portals that can buy the same kinds of vendor services. The same four matters applied to any of them that used credit-file questions: the method, the contract, the replacement and the standard.

State agencies face the trade-off with tighter budgets for fallback channels. A county office may be the in-person option for a benefit program. If online checks tighten, that office absorbs the people who fail. Its staff and hours become part of the identity system.

Banks and health systems

Banks and health systems use similar checks when customers open accounts online or register for patient portals. The stolen fields affected them through the same mechanism. A bank has an advantage NIST named: transaction history in which it took part. A health system can draw on prior visits or orders. Both can ask about relationships they control rather than about a credit file.

Federal financial and health privacy rules apply to those firms in their own terms, which we do not analyze here. In our reading, the design question is the same as the agencies’: whose secret opens the door.

Objections, in order

1. Equifax said its core credit databases were untouched, so the questions stayed secret

The release did say that. But a KBV check has two stages. The fields the release listed were the ones used to start a session. NIST required that information needed to begin KBV also be expected to be known only to the applicant and the authoritative source. In our reading, the breach weakened that condition whatever happened to the credit histories.

2. The questions were already weak, so the breach changed nothing

Identifying data had leaked before. The objection has force. Still, GAO tied its 2019 assessment of the risk explicitly to the 2017 Equifax breach. The three federal customers reviewed their processes because of it. A known weakness acquired a specific, dated reason to act.

3. Alternatives exclude vulnerable people, so KBV is the fairer method

Agency officials raised that concern with GAO, and HHS relied on it. The objection is serious. But KBV also excludes people with thin credit files. It fails people whose stolen data lets an impostor enroll first. The fair comparison is between whole journeys, including mail and in-person paths, not between one online screen and another.

4. Contract clauses cannot prevent a breach

True. Notice clauses do not stop attackers. They decide how fast a customer can act and what it can learn. GAO reported that the major federal customers learned of the breach at the public announcement. In our view, earlier notice would have let them assess their exposure before the publicity reached impostors.

5. NIST guidance already covered this, so nothing new was learned

The 2017 guideline did anticipate stolen data. What the period taught was about implementation, not principle. GAO found that guidance without implementation help or progress reporting left six agencies at four different points by May 2019.

6. Phone and document checks have their own fraud problems

They do. Each method moves risk rather than removing it. In our reading, that argues for layered checks and for watching accounts after enrollment. It does not argue for keeping a method whose inputs were exposed on a large scale.

What an identity program owner could check

The list below is a set of checks we would run today. It reflects our reading of the sources, not a regulatory requirement.

  • Which online services resolve identity using name, birth date, address and Social Security number, and what second step follows each one.
  • Whether any second step still asks questions drawn from credit files or public records, and for which applicants.
  • Whether vendor contracts require notice of a breach anywhere in the data the service depends on, including systems that do not serve you, and how fast.
  • What breach data you may receive from a vendor, in what format, for what uses and for how long.
  • Pass, fail and completion rates for each channel, including mail and in-person paths, reviewed together with fraud losses.
  • Account changes after enrollment, such as payment or address changes, tracked by the method that opened the account.
  • Whether a second source or a manual path can carry the service if a vendor must be suspended.
  • Who coordinates your response when a vendor shared with other agencies is breached, and with whom you would share assessments.

With hindsight

GAO published its account of the breach and the federal response in August 2018. It reported that attackers had gained access through the online dispute portal that held documents used to resolve consumer disputes. It said the breach exposed personal information of at least 145.5 million individuals. Equifax had raised its count from 143 million to 145.5 million on October 2, 2017.

That report documented what three of the company’s major federal customers did. IRS and SSA obtained lists of affected individuals. All three ran their own on-site reviews. IRS and SSA rewrote notice terms, IRS to a one-hour clock. USPS removed certain questions and added mailed codes. IRS stopped work on its short-term order and moved to another vendor after GAO denied the protest. GAO made no recommendations in that report.

GAO also recorded that the Bureau of Consumer Financial Protection and the Federal Trade Commission had begun investigating the breach in September 2017. It reported that DHS had offered help that the company reportedly declined.

In May 2019, GAO turned to the method. Its report on online identity verification found that agencies relied on consumer reporting agencies for knowledge-based verification. It said data stolen in breaches such as Equifax’s could be used to answer the questions. Two of six agencies, GSA and IRS, had eliminated KBV. CMS had no plans to do so, and HHS disagreed with GAO’s recommendation.

On July 22, 2019, the FTC announced a settlement with Equifax that also involved the Consumer Financial Protection Bureau and states. The FTC said Equifax had agreed to pay at least $575 million and potentially up to $700 million. The FTC, the CFPB and the states had alleged that the company’s failure to take reasonable steps to secure its network led to a breach affecting approximately 147 million people.

The FTC release, describing its complaint, said hackers stole at least 147 million names and dates of birth, and 145.5 million Social Security numbers. GAO had recorded a second revision on March 1, 2018: about 2.4 million U.S. consumers whose names and partial driver’s license information were stolen, some of them already in the 145.5 million. In our reading, the two totals describe different fields and dates, not a contradiction. The FTC alleged violations of the FTC Act and the Gramm-Leach-Bliley Act’s Safeguards Rule.

The proposed order’s security terms are the part closest to our argument. Equifax would have to run an information security program with annual risk assessments and safeguards including patch management. It would have to obtain third-party assessments every two years and annual certifications from its board or a relevant subcommittee. It would have to ensure that service providers with access to personal information maintained adequate safeguards.

The settlement also gave consumers a remedy aimed at monitoring. Starting in January 2020, Equifax was to provide all U.S. consumers six free credit reports each year for seven years, on top of the existing free annual report. Those terms addressed the consumer harm. They did not address how agencies proved identity.

That question was settled, for federal guidance, in July 2025. NIST published SP 800-63A-4, which supersedes the 2017 guideline. As checked on November 15, 2026, its text says knowledge-based verification or knowledge-based authentication shall not be used for identity verification. It allows a provider to employ KBV as part of its fraud management program.

The 2025 revision also requires providers to establish and maintain a fraud management program with identification, detection, investigation, reporting and resolution capabilities. It recommends sharing fraud events with relying parties in real time. In our reading, that is the layered approach the 2017 period pointed toward: stop relying on questions to prove identity, and watch what happens after enrollment.

What held up from the week is the diagnosis. The fields Equifax listed were the ones that open a remote identity check. GAO drew the link in 2019, and NIST closed the door on KBV for verification in 2025. The reliance on facts held by a private company proved to be a reliance on that company’s security.

What did not hold up was the expectation that guidance alone would move agencies quickly. GAO’s May 2019 count showed six agencies at four different points. The reasons they gave were cost and exclusion of parts of the public, and those reasons were real.

Some questions remain open in the sources we cite. We have not found a government-wide rule on breach notice terms for identity data vendors. We have not checked each agency’s current proofing method against the 2025 guideline. Agencies that still use history questions in any form should check them against the current text and their counsel’s reading.

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

  • identity proofing
  • knowledge-based verification
  • NIST SP 800-63
  • GAO
  • vendor contracts
  • digital services