Get Transcript’s low-risk-rated 2015 login let thieves in; IRS estimated about 22 percent of real users failed
An August 17, 2015 IRS statement reported an estimated 220,000 more attempts that cleared Get Transcript’s verification with sensitive taxpayer data. Its auditor later found the login single-factor and its low risk rating wrong; the IRS estimated about 22 percent of legitimate users failed it.
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
An August 17, 2015 IRS statement reported an estimated 220,000 more attempts that cleared Get Transcript’s verification with sensitive taxpayer data. Its auditor later found the login single-factor and its low risk rating wrong; the IRS estimated about 22 percent of legitimate users failed it.
- The week The statement first recalled the May findings.
- The gate: a multistep process with one kind of factor The IRS described its 2015 process as a multistep authentication with personal verification questions. [1] The inspector general’s November 2015 report described the steps.
- The count: attempts, accounts and a number that moved The wording was careful.
- The fallback: what taxpayers could do while the door was shut Mailing to the address of record is itself a security control.
- The rebuild: stronger proof, and who it leaves out One feature of the relaunch gave users a way to spot misuse.
- How the case applies to other public and regulated services State revenue departments, benefits programs and licensing agencies may run online services that verify identity.
- The objections, and what the record shows The IRS said exactly this: the thieves had information from sources outside the IRS before they began. [1] That is true and does not change the design question.
- Checks a service owner could run These checks are our suggestions, not requirements of any rule.
- With hindsight The inspector general reported first.
The week
On August 17, 2015, the Internal Revenue Service issued an additional statement on the incident involving its “Get Transcript” web application. [1] The application was used to obtain tax transcripts, and the agency had shut it down in May 2015. [1]
The statement first recalled the May findings. It said unauthorized third parties already had sufficient information from a source outside the agency before they accessed the application. [1] That information let them clear “a multi-step authentication process, including several personal verification questions that typically are only known by the taxpayer.” [1]
It then restated the May figures. Third parties with sensitive data from outside sources had cleared the verification process on about 114,000 total attempts. [1] Another 111,000 attempts failed the final verification step, so those attempts did not reach account information. [1]
The new part was a wider review. The agency said it had analyzed more than 23 million uses of the system over a period covering the 2015 filing season. [1] The review identified an estimated additional 220,000 attempts in which individuals with sensitive personal data cleared the verification process. [1] It identified an additional 170,000 suspected attempts that failed to clear authentication. [1]
The statement described the response. Letters would go out within days to about 220,000 people where there were instances of possible or potential access to account information. [1] Letters would also go to approximately 170,000 other households, telling them their personal information could be at risk even though identity thieves had failed to get in. [1]
The statement was open about uncertainty. It referred to “the uncertainty in many of these cases,” giving as an example a return filed before the online access occurred. [1] The letters would therefore say that recipients could disregard them if they had in fact sought their own records. [1]
The agency said it believed some of the information may have been gathered to file fraudulent returns in the 2016 filing season. [1] It encouraged recipients to use the free credit monitoring and Identity Protection PIN it was offering. [1] It said the PIN could be used to verify the authenticity of the next year’s return. [1]
It also said what it would change. The agency was working to strengthen security for the application, “including by enhancing taxpayer-identity authentication protocols.” [1] While the online service stayed down, people had other options to obtain their records. [1] The statement closed by saying the matter remained under review by the Treasury Inspector General for Tax Administration and the agency’s Criminal Investigation division. [1]
Three things in the statement framed the week for any agency or institution that serves the public online. A login built on personal questions had been passed by people holding stolen data. [1] The count of affected people depended on how the agency chose to measure, and it had grown. [1] And a service people had come to use online was closed, with slower channels left open. [1]
The statement did not say how the application’s risk had been rated before launch, or how many genuine users had failed the same questions. Those answers came later, from the inspector general, and they are the center of what follows.
The Get Transcript incident is remembered as a breach. For an agency program director, a chief information officer or the head of a digital service, it reads more usefully as a test of identity proofing at the front door of a public service. A process built on what a person knows was passed by thieves who knew it. [1] By the IRS’s own estimate, recorded later by its inspector general, about 22 percent of legitimate taxpayers could not pass it. [5]
Our argument is that the August 2015 review exposed a gate that failed in both directions, and that the fix traded one failure for more of the other. The 2015 process let in impostors with stolen data. [1] It also turned away a share of real users. [5] The stronger process the IRS launched in 2016 asked for more, and the IRS expected even more taxpayers to be unable to authenticate. [4] [5] In our reading, the citizen experience question was never whether to secure the service. It was what the people who could not pass the gate would do instead.
That framing matters for leaders who own service channels. Authentication is designed by security teams and measured by fraud rates. Its failures to admit real users show up as calls, mailed requests and lost trust. The IRS’s choices in 2015 and 2016 made both kinds of failure visible. [4] [5]
We take the incident in four parts:
- The gate: what the 2015 process asked of users, and why the auditor said it was single-factor.
- The count: what the August review measured, and why the number kept moving.
- The fallback: what taxpayers could do while the online service was down.
- The rebuild: what the stronger process asked for, and who it would leave behind.
The parts do not overlap. The first is design, the second measurement, the third continuity and the fourth redesign. After them we look at how the case applies to other public and regulated services, take the objections in order and end with checks a service owner could run.
The gate: a multistep process with one kind of factor
What the process asked
The IRS described its 2015 process as a multistep authentication with personal verification questions. [1] The inspector general’s November 2015 report described the steps. It said the IRS asked for basic identifying information and required users to answer knowledge-based questions provided by a third-party credit reporting agency. [2] It said users also gave an email address, to which the IRS sent a confirmation code. [2]
The report said those steps did not add up to multifactor authentication. The confirmation code was not sent to an email address on the taxpayer’s record, and was not a code that served as a second factor proving identity. [2] Its conclusion was that the methods provided only single-factor authentication, “despite NIST standards requiring multifactor authentication for such high-risk applications.” [2]
Why the steps were one factor
The distinction is between steps and factors. A factor is a kind of evidence: something a person knows, has or is. The September 2016 report gave that three-part definition. [5] Basic identifying information and answers to credit history questions are things a person knows, and things a thief can learn. [2] An email address chosen by the user at sign-up proves only that the user controls some mailbox. [2]
Adding more steps of the same kind adds friction without adding a different kind of proof. In our reading, that was the design flaw the incident exposed. Each step could be passed by anyone holding the same stolen file.
The standards the inspector general applied
The November 2015 report measured the process against federal requirements. It said OMB standards require agencies to assess the risk of authentication error for each online service. [2] It defined an authentication error as an agency confirming an identity when the person is not who they claim to be. [2] It said NIST Special Publication 800-63 sets the requirements an agency’s process must meet for a given level of assurance. [2]
That structure links two decisions. The risk assessment sets the required assurance level, and the level sets the required evidence. A low rating permits weaker evidence. [2] In our reading, that is why the inspector general focused on the rating: once it was set too low, the rest of the design followed correctly from a wrong premise.
The questions came from a credit bureau
The knowledge-based questions were provided by a third-party credit reporting agency. [2] That arrangement meant the agency’s gate depended on data held by a private company and drawn from consumers’ credit files. The questions were only as secret as those files.
In our reading, that dependency is a vendor risk as well as a design choice. A service owner who buys knowledge-based questions buys the assumption that their answers are not widely known. When large stores of personal data are stolen elsewhere, the assumption fails for every customer of the vendor at once, and no contract term can restore it.
How the risk was rated
The November 2015 report found that the IRS had assessed the authentication risk of Get Transcript as required, but had rated it low to both the IRS and taxpayers. [2] A low rating meant the likelihood of an impostor obtaining and using the information was low and controls would impede impostors. [2] The report said the IRS “now knows” the risk was high and should have required multifactor authentication. [2]
This is the governance lesson inside the technical one. The level of authentication followed from a risk rating. The rating underestimated what a transcript was worth to a criminal. [2] In our reading, a transcript held prior-year income data that could help file a convincing fraudulent return, which the IRS itself said it feared. [1]
Burden as a reason for a weaker gate
The November 2015 report described a related decision for the IP PIN application. It said the IRS had not completed an authentication risk assessment for that application. [2] It said management recognized the application required multifactor authentication but believed requiring it would further burden identity theft victims trying to obtain a PIN. [2] The inspector general said it believed a risk assessment would have found the risk of compromise outweighed the burden. [2]
That exchange states the trade-off this article is about in its plainest form. A protection meant for victims was made easier to obtain so victims could get it, and that ease also helped impostors. In our reading, the decision was defensible as a service goal and weak as a security one, and the missing risk assessment meant the trade was never written down.
How the attack was noticed
The September 2016 report described detection. On May 14, 2015, the IRS Computer Security Incident Response Center identified a significant number of undeliverable emails sent by the authentication application. [5] They were the confirmation code emails, and their suspicious characteristics led to a security review. [5]
The same report found weaknesses in monitoring. It said audit log reports were not adequately monitored at the time. [5] Its example was a user who, in July 2014, attempted to authenticate 902 times within one 24-hour period, far beyond the unusual activity trigger. [5] It said poor communication between the IRS and its contractor meant the IRS did not fully know what was being screened at its Integrated Enterprise Portal. [5]
What the gate part shows
The gate failed against attackers who had the answers, and its monitoring did not stop automated attempts. [2] [5] In our reading, the IRS found the attack through a side effect, bounced emails, rather than through controls built to detect it. [5] For a service owner, the point is that authentication strength and attack detection are separate controls. Both were weak.
The count: attempts, accounts and a number that moved
What was counted
The August statement counted attempts that cleared verification, not confirmed victims. [1] The IRS said the new review identified an estimated 220,000 attempts where individuals with sensitive data cleared the process. [1] It sent letters to about 220,000 taxpayers where there was possible or potential access. [1]
The wording was careful. “Possible or potential access” is not a finding that every letter recipient was a victim. The IRS said some recipients had filed a return before the access occurred, and invited them to disregard the letter if they had sought their own records. [1]
Uses, transcripts, attempts and accounts
The documents used different units. The August statement counted “uses” of the system, more than 23 million in its review, and “attempts” that cleared or failed verification. [1] The February 2016 statement said approximately 23 million transcripts were ordered in the 2015 filing season, and counted taxpayer accounts with potential access. [3] The September 2016 report said the IRS provided 23 million transcripts through the application from October 1, 2014, through April 15, 2015, and estimated potentially affected accounts. [5]
Those units are not interchangeable. One person can make several attempts, one attempt can involve several transcripts, and one email address can be used for several accounts. [3] In our reading, a service owner reporting an incident should state the unit of every figure, because a reader comparing an attempt count with an account count will draw the wrong conclusion.
Why the number grew
In May, the IRS had counted about 114,000 attempts that cleared verification. [1] The August review covered the 2015 filing season and more than 23 million uses. [1] In February 2016, the IRS said a nine-month review by the inspector general, going back to the application’s launch in January 2014, found potential access to approximately 390,000 additional taxpayer accounts. [3] It found 295,000 more transcripts that were targeted without success. [3]
Each expansion came from a wider window or a different detection method. The inspector general’s investigators identified suspicious email addresses that made multiple attempts to access accounts. [3] The IRS noted that some of those might be family members, tax preparers or financial institutions using one email address for several accounts, and said it would notify everyone affected anyway. [3]
What a moving number does to a service
A count that keeps growing is a communications problem as well as a security one. Each new mailing reaches people who had been told nothing before, long after the access. [1] [3] Each one also adds call volume, identity theft affidavits and requests for protection products.
In our reading, the IRS chose breadth over precision: it notified anyone who met its criteria rather than wait for certainty. [3] That choice protects people who were affected. It also sends alarming letters to people who were not, which the IRS acknowledged by telling them to disregard the letter if appropriate. [1]
The harm the count stood for
The September 2016 report put figures on the downstream fraud. It said the IRS, with the inspector general’s help, had estimated the total potential unauthorized accesses at 724,000 taxpayer accounts. [5] The IRS had identified approximately 252,400 potentially fraudulent returns related to the incident. [5] It stopped approximately 189,400 returns claiming $1.55 billion in refunds, and had issued refunds on 63,000 returns totaling $490 million. [5]
The report recorded the IRS’s caution that its analysis was ongoing and that some apparently unauthorized accesses might prove legitimate. [5] Those qualifiers belong with the figures. They describe an estimate, not a closed ledger.
Letters as a service channel
A notification letter may be the first contact an affected person has with an incident, and its design shapes what they do next. The August letters told recipients the agency was offering credit monitoring and a PIN, and that they could disregard the letter if they had made the request themselves. [1] Those two messages ask the reader to make a judgment: was this me?
In our reading, letters of that kind work best when they give the date of the access in question and a clear way to confirm or deny it. The statement did not describe the letters’ content in that detail. [1] A service owner designing such letters would test them with people who are not security specialists before mailing hundreds of thousands.
What the count part shows
A service owner who must notify people after an authentication failure faces a choice of method. A narrow count is defensible but may miss victims. A wide count reaches more victims but alarms people who were not affected. [1] [3] The IRS widened its count more than once. In our reading, the lesson for others is to decide in advance how such a count will be made, so that the first number published is not the smallest.
The fallback: what taxpayers could do while the door was shut
Transcripts by mail
When the IRS took Get Transcript offline in May 2015, it kept other routes open. [1] The September 2016 report said taxpayers could still order a transcript online and have it mailed to their address of record. [5] The February 2016 statement said online requests were being taken for mailed copies, which could typically take five to 10 days to arrive. [3]
Mailing to the address of record is itself a security control. It relies on the IRS’s own record of where the taxpayer lives, not on information the requester supplies. In our reading, that is why the mail channel stayed open while the online one closed: it had a factor the online process lacked, possession of a mailbox the IRS already knew.
Who relies on transcripts
The IRS’s June 2016 release explained why the service mattered. It said transcripts are used for non-tax purposes such as income validation for mortgages or student loans. [4] It said taxpayers also use them to obtain prior-year adjusted gross income, which they need to e-file. [4]
That makes the transcript a dependency for other institutions. A lender or a student aid office that asks an applicant for a transcript relies on the applicant’s ability to get one. When the online channel closed, that dependency slowed to the speed of mail. [3] [4]
The cost of the slow channel
A mailed record that takes five to 10 days to arrive is a different service from an instant download. [3] For a person applying for a mortgage or student aid, that delay can move a closing date or a deadline. The June 2016 release named both uses. [4]
That cost falls on the people who fail or avoid online proofing, and on the institutions waiting for their documents. In our reading, a service owner should know which downstream processes depend on its records and how long they can wait. That knowledge decides whether a mail fallback is adequate or whether an assisted channel, such as an office visit or a phone call with stronger checks, is needed.
Protection for those affected
The fallback also included protection. The August statement offered free credit monitoring and Identity Protection PINs to letter recipients. [1] The February 2016 statement said taxpayers whose transcripts were accessed could request an IP PIN by filing Form 14039, the Identity Theft Affidavit. [3] It said the IRS was offering a free identity theft protection product for one year and placing special markers on affected accounts to alert IRS assistors. [3]
Those markers are a service design choice. A taxpayer who calls about a letter reaches an assistor who can see that the account was part of the event. [3] In our reading, that turns a confusing call into a recognized case, which is the experience a frightened caller needs.
Planning for a closure
The agency could not have planned the 2015 closure in detail, but the general case is foreseeable: an online service may have to be switched off at short notice. [1] When that happens, demand moves to the remaining channels in a single step. In our reading, the capacity of those channels, the scripts assistors use and the messages on the closed service’s page are things that can be prepared in advance and tested once a year.
Sharing with partners
The February 2016 statement said the IRS was sharing information about the incident with the states through the Security Summit, a partnership of the IRS, state revenue departments and the tax industry. [3] The inspector general’s November 2015 report described the Summit’s working groups, including one on authentication. [2] An agency’s authentication failure is a risk to its partners, because the same stolen data can be used against their systems.
What the fallback part shows
The IRS could close its online door because other doors existed. [3] [5] In our reading, the existence and capacity of those other doors decided how much harm the closure caused to legitimate users. An agency that has retired its paper and phone channels has fewer choices when an online service must close.
The rebuild: stronger proof, and who it leaves out
What the 2016 process required
On June 7, 2016, the IRS relaunched Get Transcript Online with a new authentication framework. [4] The release said it was built with the assistance of the U.S. Digital Service and other security authorities, and met NIST and OMB security standards. [4] The framework required a two-step authentication process for online tools that require a high level of assurance. [4]
To register, a taxpayer needed an email address, a text-enabled mobile phone and specific financial account information, such as a credit card number or certain loan numbers. [4] Codes would be sent by email and text, and returning users would always have to enter a text code. [4] Users who had registered under the old process had to re-register. [4] The service would show when the account was last accessed. [4]
Showing users their own history
One feature of the relaunch gave users a way to spot misuse. The service would show the date and time the page was last accessed. [4] That puts a detection control in the user’s hands. A person who sees an access they did not make has a reason to call. In our reading, it is a low-cost addition to any service that releases sensitive records, and it turns users into a second line of monitoring.
The trade the IRS stated
The IRS did not present the change as free. The release said some taxpayers may find it more difficult to authenticate under the strengthened process. [4] The September 2016 report recorded that IRS management had said it must balance stronger authentication against legitimate taxpayers’ access. [5] It recorded the estimate that about 22 percent of legitimate taxpayers could not authenticate under the single-factor process, and the IRS’s expectation that an even higher percentage would fail under multifactor authentication. [5]
The report also recorded the safety valve. All taxpayers would continue to be able to order a transcript online or by telephone and have it mailed to their address of record. [5] In our reading, that sentence is what made the stricter gate acceptable: no one lost access to their records, only the speed of access.
Who a phone and a financial account leave behind
The 2016 requirements assumed that a taxpayer had a text-enabled mobile phone and a credit card or loan account. [4] A taxpayer without one of those could not complete online registration. The IRS’s answer was the mail channel. [4] [5]
For a public service, that is a design choice with equity effects. A taxpayer without a credit card or loan account, or without a text-enabled phone, would depend on the slower channel. The rebuild did not create that gap, but it formalized it. In our reading, a public service that adds stronger proofing should measure who fails it as well as how many.
Detection as well as proof
The September 2016 report showed that the rebuild reached beyond the front door. It recommended clarifying IRS and contractor responsibilities for preventing automated attacks, monitoring audit trails, giving security specialists tools and training, and recording which application a user intended to reach. [5] The IRS agreed with all seven recommendations. [5] Those controls address the volume and pattern of attempts, which the 2015 design had missed. [5]
Contractors in the authentication chain
The September 2016 report found that the IRS did not clearly specify which parties, including IRS divisions and contractors, were responsible for detecting and preventing automated attacks. [5] It said poor communication with the contractor running the Integrated Enterprise Portal left the IRS unaware of what the portal screened. [5]
That finding applies to any service whose front door is operated by a vendor. The agency owns the risk, while the vendor operates the controls. In our reading, a contract or service agreement that does not name who watches for automated attempts, and who acts on them, leaves a gap that neither side may notice until an attack exploits it.
From self-service to assisted proofing
The 2016 rebuild kept proofing self-service, with stronger evidence. [4] By 2022, the agency added a different route: a live, virtual interview with agents for people who did not want to use biometric verification. [7] That option put a person back into the proofing process for those who chose it.
In our reading, an assisted route is the counterpart to a strict automated one. It costs staff time per user. It also serves people whose records do not fit the automated checks, which is the group the 2015 and 2016 estimates suggested was large. [5]
What the rebuild part shows
The rebuild added factors the 2015 process lacked: possession of a phone and knowledge of a financial account number tied to the person. [4] It accepted that more people would fail and relied on mail for them. [5] The general lesson for service owners is that identity proofing and inclusion are designed together. A stronger gate needs a staffed alternative.
How the case applies to other public and regulated services
State and local agencies
State revenue departments, benefits programs and licensing agencies may run online services that verify identity. The IRS’s inclusion of states in the Security Summit reflects that the same stolen data can be used across tax systems. [3] In our reading, any agency using knowledge-based questions from credit data for a service that releases sensitive records faced the same weakness the inspector general described. [2]
Banks and credit unions
A financial institution may use knowledge questions in account opening and account recovery. The Get Transcript case is about a government service, and the standards the inspector general cited were federal ones. [2] For a bank, the relevant expectations come from its own regulators. The mechanism is the same, though: a question whose answer is in a stolen file is a weak factor wherever it is asked.
Health systems
Patient portals release health records. A health system that verifies new portal users with knowledge questions relies on the same kind of evidence. [2] The fallback lesson applies too. A patient who fails online proofing needs a route through the registration desk or a mailed code to the address on file.
Federal agencies
Federal agencies were bound by OMB and NIST guidance on authentication assurance, which the inspector general applied to the IRS. [2] NIST’s current identity proofing guideline, published in July 2025, states that knowledge-based verification “SHALL NOT be used for identity verification.” [8] For a federal service owner, the 2015 weakness is now addressed directly in the standard.
Vendors that sell verification
Agencies and institutions buy identity verification from vendors, as the IRS did for its knowledge questions in 2015 and for biometric verification later. [2] [6] The case suggests questions for those purchases: what kinds of evidence the vendor uses, how its failure rates are measured and reported by population, and what happens to data collected during proofing. The IRS’s 2022 statement said images provided for biometric verification would be deleted, and existing biometric data permanently deleted. [7] In our reading, those terms belong in the contract from the start.
The objections, and what the record shows
“The IRS was the victim of data stolen elsewhere”
The IRS said exactly this: the thieves had information from sources outside the IRS before they began. [1] That is true and does not change the design question. A process that relies on information available outside the agency will be passed by anyone who has that information. [2] The inspector general found the risk rating, not the theft elsewhere, to be the decision the IRS controlled. [2]
“Stronger proofing harms the people the service is for”
The IRS’s own estimates support part of this. About 22 percent of legitimate users failed the old process, and more were expected to fail the new one. [5] The answer the IRS gave was a fallback, not a weaker gate. [5] The objection is strongest where no fallback exists. Where one does, it becomes a question of the fallback’s speed and capacity.
“Counting attempts overstates the harm”
It may. The IRS said some letter recipients had made the requests themselves, and later noted shared email addresses might belong to families or preparers. [1] [3] The alternative, counting only confirmed misuse, would have left victims uninformed. The fraudulent return figures later reported show the harm was real for a large number of returns. [5]
“Monitoring would have caught it”
The inspector general’s monitoring findings support this partly. A user making 902 attempts in a day exceeded the unusual activity trigger, and audit logs were not adequately monitored. [5] Better monitoring might have shortened the attack. It would not have fixed a gate that answered correctly to anyone with the right data. Both controls were needed. [2] [5]
“The answer was obvious: require multifactor from the start”
With hindsight it looks that way, and the inspector general said multifactor authentication should have been required. [2] The record also shows why it was not. The IRS’s framework was designed to offer variable assurance levels, and the capability for multifactor authentication was still being built. [2] [5] The IRS also faced the access problem directly: it expected more legitimate users to fail a stronger process. [5] The decision that went wrong was the rating that made single-factor seem acceptable, not ignorance that multifactor existed. [2] For other service owners, the useful step is to revisit ratings when the threat changes, rather than only when a service launches.
“Fallback channels are attack paths too”
This is a fair challenge to our argument. A mail channel depends on the accuracy of the address on file, and a phone channel depends on the checks an assistor applies. In our reading, the mail fallback in this case was safer than the online gate because it sent records to an address the agency already held, rather than to one the requester supplied. [5] A fallback that releases records to new contact details would not share that property, and would need its own controls.
Checks a service owner could run
These checks are our suggestions, not requirements of any rule. Agencies and institutions should apply the standards that bind them and confirm legal questions with counsel.
- Factor inventory: for each online service that releases sensitive records, list the kinds of evidence used and mark any that a person’s stolen file would contain.
- Risk rating review: re-rate each service by what its data is worth to a criminal as well as to the user, and record the reasoning.
- Failure rate: measure the share of legitimate users who fail proofing, and who they are, alongside the fraud rate.
- Fallback capacity: confirm that a mail, phone or in-person route exists for each online service, and that it can absorb demand if the online service closes.
- Attempt monitoring: define who watches authentication volumes and repeated attempts, including any contractor, and test that alerts reach someone.
- Notification method: decide in advance how affected users will be counted and notified after an authentication failure.
- Assistor support: make sure front-line staff can see when a caller’s account was part of an incident.
- Partner sharing: agree how incident details will be shared with partners who verify the same people.
The August 2015 statement showed a gate that thieves passed and that the IRS would rebuild. [1] The checks above follow both sides of that rebuild: make the gate harder for impostors, and make sure the people it stops have another way in.
With hindsight
The inspector general reported first. On November 19, 2015, TIGTA issued a report finding that authentication methods for the IRS’s online services did not comply with government information security standards. [2] It said the Get Transcript and IP PIN applications provided only single-factor authentication. [2] It recommended a Service-wide authentication strategy, and the IRS said it had created a new executive position to lead it. [2]
The count grew again. On February 26, 2016, the IRS said an inspector general review back to January 2014 had found potential access to approximately 390,000 additional taxpayer accounts, and 295,000 more transcripts targeted without success. [3] Mailings were to start on February 29. [3] The statement said 47 million transcripts had been ordered through the tool since its launch in 2014. [3]
The service came back with stronger proofing. On June 7, 2016, the IRS relaunched Get Transcript Online with a process requiring an email address, a text-enabled mobile phone and specific financial account information. [4] It said some taxpayers might find the process more difficult, and kept options to receive transcripts by mail. [4]
On September 7, 2016, TIGTA reported on the response. It said the IRS’s initial incident handling generally followed federal guidance. [5] It found weaknesses in detecting automated attacks and monitoring audit logs, and made seven recommendations, all agreed. [5] It recorded the IRS’s estimate of 724,000 potentially affected accounts and the fraudulent return figures, with the IRS’s caution that analysis was ongoing. [5]
The trade between security and access returned in 2022. On February 7, 2022, the IRS announced it would transition away from using a third-party service for facial recognition to help authenticate people creating new online accounts. [6] On February 21, 2022, it announced a new option to verify identity during a live, virtual interview with agents, without biometric data. [7] It said it would work with the General Services Administration toward offering Login.gov. [7]
The standard moved too. NIST’s SP 800-63A-4, published in July 2025, states that knowledge-based verification or knowledge-based authentication “SHALL NOT be used for identity verification.” [8] That is the kind of evidence the 2015 process relied on. [2]
The November 2015 report had also recorded the timetable. It said the IRS did not expect to have multifactor capability in place before the summer of 2016. [2] The June 2016 relaunch met that expectation. [4]
What held up from August 2015 is the IRS’s diagnosis that the attackers came prepared with data from elsewhere. [1] The fix that followed did not try to make that data secret again. It added possession of a phone and a second channel, and later live verification by people. [4] [7]
What did not hold up was the initial count. The number of affected accounts the IRS published in August 2015 was revised upward in February 2016, and the September 2016 report put the estimated total at 724,000 taxpayer accounts. [1] [3] [5]
In our view, the lasting point for any public or regulated service is the pairing the IRS arrived at. Stronger proof at the online gate, and a slower, staffed path for those who cannot pass it, were designed as one system.
Spotted an error? Tell us and we will check it and correct the article where needed.