The 2015 FREEDOM Act kept call records at the carriers; data errors there later helped end the program

The USA FREEDOM Act, approved June 2, 2015, ended bulk collection and had the government query records that phone companies kept. A 2020 oversight report found that provider data errors contributed to NSA’s decisions to delete the data and suspend the program.

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 USA FREEDOM Act, approved June 2, 2015, ended bulk collection and had the government query records that phone companies kept. A 2020 oversight report found that provider data errors contributed to NSA’s decisions to delete the data and suspend the program.

  1. The week The law’s legislative history, printed at its end, gives the route.
  2. The query: what the government could ask for The law also listed what would not qualify on its own.
  3. The holder: what stayed with the companies The second part concerns the companies that held the records.
  4. The record: what the checks could and could not see The third part is the quality of what came back.
  5. The disclosure: what others could learn about the requests The fourth part is disclosure.
  6. How it reached different organizations For an agency, the program is a case study in dependence on external records.
  7. Objections, and where the argument stops The first objection is that the program’s end was a policy decision about intelligence value.
  8. Checks for leaders who build on others’ records These are our own suggestions, drawn from the program’s history.
  9. With hindsight The law’s call record program started later in 2015.

The week

On June 2, 2015, the USA FREEDOM Act of 2015 became law as Public Law 114-23. [1] Its full name is the Uniting and Strengthening America by Fulfilling Rights and Ensuring Effective Discipline Over Monitoring Act of 2015. [1] The enrolled text records that it was approved on June 2, 2015. [1]

The law’s legislative history, printed at its end, gives the route. The House considered and passed H.R. 2048 on May 13. [1] The Senate considered and passed it on May 31, June 1 and June 2. [1] The text itself shows the deadline Congress had been working against. Section 705 struck the date “June 1, 2015” from two earlier sunset provisions and inserted “December 15, 2019.” [1] Those provisions set when certain surveillance authorities, including the business records authority at 50 U.S.C. 1861, would stop running. [1]

The center of the law was a set of limits on bulk collection. Section 103 required each application for business records to name “a specific selection term to be used as the basis for the production.” [1] It also said no order could authorize collection “without the use of a specific selection term” meeting those requirements. [1] Section 107 defined the term. Outside the call record track, it was a term that “specifically identifies a person, account, address, or personal device, or any other specific identifier.” [1] It excluded an identifier for a broad geographic region, such as a city, a state, a zip code or an area code, when not used as part of a specific identifier. [1]

Section 101 set a separate track for call detail records produced on an ongoing basis. [1] An application had to show reasonable grounds to believe the records were relevant to an authorized international terrorism investigation. [1] It also had to show a “reasonable, articulable suspicion” that the selection term was associated with a foreign power engaged in international terrorism, or its agent. [1] An order could authorize daily production for a period not to exceed 180 days. [1] It could allow a first set of records based on the selection term and a second set based on identifiers found in the first. [1]

The law defined a call detail record narrowly. It covered session-identifying information, such as an originating or terminating telephone number, a calling card number, or the time or duration of a call. [1] It excluded the contents of any communication; the name, address or financial information of a subscriber or customer; and cell site location or GPS information. [1]

The records stayed with the companies that created them. Orders were to provide that records be produced “in a form that will be useful to the Government.” [1] Companies were to give the government the information, facilities or technical assistance needed, while protecting secrecy and minimizing interference with service. [1] The government had to compensate them for reasonable expenses, and the law barred lawsuits against those who complied. [1] The government was to adopt minimization procedures requiring prompt destruction of records it determined were not foreign intelligence information. [1]

The law reached beyond phone companies. Section 501 required national security letters under the Right to Financial Privacy Act to include a term specifically identifying “a customer, entity, or account.” [1] It made similar changes to letters under the Fair Credit Reporting Act and to letters for telephone toll and transactional records. [1] Section 603 let recipients of these orders and letters publish aggregate counts in set bands. [1] The amendments in sections 101 through 103 were to take effect “180 days after the date of the enactment of this Act.” [1]

The USA FREEDOM Act is remembered as the law that ended the government’s bulk collection of phone records. That is accurate as far as it goes. For a technology or data leader, its more durable feature was a design choice. The law did not tell the government to stop using call records. It told the government to stop holding them in bulk and to ask the companies that held them, one specific term at a time. [1]

That choice turned an intelligence program into a data supply chain. The records stayed in carriers’ systems, built for billing and network operations, and reached the government in response to court-approved queries. [1] [2] The Privacy and Civil Liberties Oversight Board later reported that this arrangement ran into data-integrity problems in the providers’ records. [2] It found those problems contributed to the National Security Agency’s decisions to delete the data and to suspend the program. [2]

Our argument is that the law’s architecture moved the hardest problem from collection to provenance. Once the data stayed with others, the government’s ability to use it depended on the accuracy of records it did not create and could not fully check. In our reading, that lesson applies to any agency, bank or health system that builds a service on records held by someone else.

We take the law in four parts, which do not overlap:

  1. The query: what the government could ask for, and how narrowly.
  2. The holder: what stayed with the companies, including banks and credit bureaus.
  3. The record: how the data was checked, and what the checks could not see.
  4. The disclosure: what the public, the courts and recipients could learn about the requests.

The first part is the request, the second the party that answers it, the third the quality of the answer and the fourth the reporting around all three. After them we look at how the law reached different kinds of organizations, take the strongest objections, and end with checks a data or compliance leader could run.

The query: what the government could ask for

A term instead of a database

The law’s first move was to require a “specific selection term” for every business records order. [1] No order could authorize collection without one. [1] The definition named persons, accounts, addresses and personal devices, or any other specific identifier. [1] It required the term to limit the scope of what was sought “to the greatest extent reasonably practicable.” [1]

The law also listed what would not qualify on its own. An identifier for an electronic communication service provider did not qualify unless the provider was itself a subject of the investigation. [1] An identifier for a broad geographic region did not qualify either, and the text named examples: the United States, a city, a county, a state, a zip code or an area code. [1] Both exclusions applied when the identifier was not used as part of a specific identifier. [1]

In our reading, the exclusions were written against particular ways a broad request could be dressed as a narrow one. Naming a provider would reach all of its customers. Naming an area code would reach everyone in it. The text closed both routes while allowing multiple terms to be used together. [1]

The call record track

Call detail records had their own rules. The law created a separate application for their production “on an ongoing basis,” tied to an authorized investigation to protect against international terrorism. [1] For that track the selection term had to identify “an individual, account, or personal device.” [1] The government had to show reasonable grounds to believe the records were relevant and a reasonable, articulable suspicion that the term was associated with a foreign power engaged in international terrorism, or its agent. [1]

An order on this track could authorize daily production for no more than 180 days, with extensions on a new application and finding. [1] It allowed two rounds. The first set of records came from the approved term. [1] The second set came from session-identifying information or a calling card number identified by the term in the first round. [1] The oversight board later summarized this as collection within two hops of a specific selection term: a person’s contacts, and those contacts’ contacts. [2]

What the two-hop design meant for data

The two-hop structure matters for the rest of this piece. A second-hop request depended on the content of a first-hop record. If the first-hop record was wrong, the second-hop request went to the wrong numbers. In our reading, the design made the accuracy of the first answer a condition of the lawfulness of the second question.

The oversight board described exactly that pattern in 2020. It said that in most data-integrity cases, NSA systems unknowingly relied on inaccurate first-hop data to decide which second-hop requests to issue. [2] We return to this under the third part.

An emergency path

The law also gave the Attorney General an emergency authority. The Attorney General could require production before an order was obtained if he or she reasonably determined that an emergency required it and that a factual basis for an order existed. [1] The emergency path kept the same structure: a later application to the court, under the same standards. [1] For a records holder, it meant some requests could arrive before a court order existed, with the order to follow.

What the query part asked of the government

For the agencies involved, the first part changed the unit of work. The government no longer worked from a database it held. It worked from a list of approved terms and a stream of responses. Each term had a legal basis, a court order and an expiry date. Each response had to be traceable to the term and order that produced it.

That is a familiar problem in any regulated data operation. A bank answering a subpoena, a health system releasing records under an authorization and an agency sharing data under an interconnection agreement all need the same chain: request, authority, scope, response and retention. The 2015 law wrote that chain into statute for one category of records.

The holder: what stayed with the companies

Records kept for other purposes

The second part concerns the companies that held the records. The law did not create a new record type. It drew on records the carriers already kept for their own business. [1] The oversight board recounted the Director of National Intelligence’s explanation that NSA halted the program after balancing its intelligence value, its costs, and compliance and data-integrity concerns. [2] In that explanation, as the board quoted it, the concerns arose from the complexities of using business records that the providers themselves generated for intelligence purposes. [2]

Business records are created to run a business. A carrier’s call records serve billing, routing, interconnection and fraud control. Their fields, formats and error tolerances are set by those uses. In our reading, the law asked records built for one purpose to carry the weight of another, where a single wrong field could send a request to the wrong person.

Duties the law placed on holders

The text set out what the holders owed. Orders on the call record track had to provide that records be produced “in a form that will be useful to the Government.” [1] Each company directed to produce had to furnish all information, facilities or technical assistance necessary, protecting the secrecy of the production. [1] It also had to cause “a minimum of interference” with the services it provided. [1]

In return, the law gave holders three things. The government had to compensate them for reasonable expenses of production and technical assistance. [1] No cause of action could lie against a person who produced records or gave technical assistance under an order. [1] And production was not a waiver of any privilege in another proceeding. [1]

The format the holders produced

The oversight board described how this worked once the program began. NSA sent selection terms and direct contacts to the providers through a system the report calls System 1. [2] The providers searched for responsive records and produced them in a standardized format of about 50 fields per record. [2] The fields included the participants’ phone numbers, device identifiers where applicable, and the date, time and duration of the call. [2] Each record also carried a code indicating the specific court order. [2]

In our reading, the law’s design became an integration project here. Several providers, each with its own systems, mapped their records into a common format for a single recipient. Every difference in how a provider stored a field was a chance for a record to mean something other than what the recipient assumed.

Banks and credit bureaus

The holder part also reached financial firms. Section 501 changed the Right to Financial Privacy Act provision under which the FBI may request financial records for certain intelligence purposes. [1] Such a request now had to include “a term that specifically identifies a customer, entity, or account to be used as the basis for the production and disclosure of financial records.” [1]

The same section amended two Fair Credit Reporting Act provisions under which consumer reporting agencies can be asked for information. [1] Requests and orders under those provisions now had to include a term that specifically identifies a consumer or account. [1] Letters for telephone toll and transactional records had to use a term that specifically identifies a person, entity, telephone number or account. [1]

For a bank’s legal and records teams, the change was concrete. A national security letter that did not name a customer, entity or account would not, in our reading, meet the amended text. The team receiving it had a basis to check scope before producing anything. Banks should check with counsel before relying on this reading; the law states the requirement, not the handling procedure.

Secrecy and its review

Section 502 rewrote the nondisclosure rules for national security letters. [1] For letters to communications providers, nondisclosure applied when the FBI Director, or a designee no lower than a Deputy Assistant Director at headquarters or a Special Agent in Charge of a field office, certified that the absence of a prohibition may result in harms the law listed. [1] The listed harms included a danger to national security and interference with an investigation. [1] Recipients could seek judicial review, and the letter had to include notice of that availability. [1] The same structure applied to letters under the Right to Financial Privacy Act. [1]

The law also required the Attorney General, not later than “180 days after the date of enactment of this Act,” to adopt termination procedures. [1] They were to require review “at appropriate intervals” of whether the facts supporting nondisclosure continued to exist. [1] They also required termination when the facts no longer supported nondisclosure, and notice to the recipient. [1]

In our reading, the review procedures gave recipients a reason to keep their own records of when each letter arrived, what it covered and whether its secrecy was later lifted. A firm that cannot answer those questions for its own files cannot tell whether a notice of termination applies to it.

What the holder part cost the holders

Compensation for reasonable expenses did not remove the operating work. Someone had to receive each request, check its term and order, run the search, map the results into the required format, and deliver them securely. Someone had to answer questions when the recipient found an anomaly. The law placed that work on the companies and paid for it; it did not specify how it was to be done. [1]

The record: what the checks could and could not see

Validation of form

The third part is the quality of what came back. The oversight board described NSA’s validation step in detail. System 1 checked the order code on each record to ensure collection occurred under a valid order. [2] It checked that fields were plausible: that a date field held a valid date, and that other fields had the expected number of digits or format. [2] Records that failed were held for review by technical staff. [2]

The board was precise about the limit. This step did not allow NSA to verify that a record accurately described a phone call that had in fact occurred, or that the data did not contain errors. [2] It was designed to catch records that could not, on their face, be a valid response. [2]

In our reading, this is the distinction that decided the program’s fate. Format validation catches malformed records. It cannot catch a well-formed record that is wrong. A recipient that does not hold the source system has no way to tell the two apart without the holder’s help.

Errors in the source

The board reported that, beginning in 2016, NSA identified a series of data-integrity problems and other data errors. [2] In one incident, a provider produced records with inaccurate data, populated by the provider’s own production system. [2] The board said those inaccurate records were created over a two-year period. [2] A footnote said the number reflected a fraction of one percent of the overall collection. [2]

The board described the response. The same day it identified the problem, NSA stopped issuing new requests to that provider and stopped processing its data into repositories. [2] It warned analysts not to rely on records from the affected period. [2] The provider implemented a technical fix to prevent delivery of inaccurate records. [2]

NSA’s internal investigations then found that some inaccurate records had been used to support four applications to the surveillance court. [2] On April 11, 2018, the government notified the court of the inaccurate information and of the four applications. [2] NSA deleted the records acquired under those applications and recalled one disseminated intelligence report. [2]

When the holder cannot find the errors

The decisive finding came next. NSA searched for similar anomalies from other providers and found questionable records. [2] One provider confirmed it had also produced inaccurate records in certain situations, and had reported a separate set in which fields had been overwritten with unrelated data. [2]

By May 2018, the board said, NSA realized the providers could not identify all the affected records. [2] NSA had no way to determine independently which records were inaccurate. [2] It therefore had no viable way to remove the bad records and keep the good ones, and it began deleting all data the providers had produced under the law. [2]

In our reading, this is the central lesson of the program for anyone who builds on others’ records. The failure was not that errors occurred. It was that neither the holder nor the recipient could scope them. Without that, the only safe remedy was to discard everything.

Errors that compound

The board explained why errors here were more serious than in a single lookup. In most of these cases, the board said, NSA systems relied on inaccurate first-hop data to decide which second-hop requests to issue. [2] The result was requests for records the court had not authorized. Other incidents arose from fields overwritten with incorrect or unrelated data. [2]

Later in 2018, NSA noticed a larger than expected number of values in specific fields from one provider. [2] The provider found it had produced more records with incorrect data, and NSA could not rectify the problem with that provider. [2] The board said further discussion gave NSA a better understanding of how all providers maintained their records. [2] The government then told the court its position on why certain records remained valid session-identifying information, and the court did not explicitly address the issue. [2]

The board’s assessment of fault

The board’s findings on intent should be stated as it stated them. It found no abuse of the program and no instance of officials intentionally seeking records they knew were prohibited. [2] It found the compliance incidents were inadvertent, not willful. [2] It said NSA took steps to remedy each incident, including notifying oversight bodies, limiting requests and deleting erroneously obtained data. [2] It also found that NSA chose the narrower reading of its authority each time the scope of collection was in question. [2]

The board also recorded that some incidents could have arisen under other authorities. Its examples were information omitted from a court application, officers with access but without required training, and a provider producing data beyond an order’s end date. [2] Others, it said, raised questions unique to this law’s structure. [2]

Deletion as a data operation

The law treated destruction as part of the design. Orders on the call record track had to direct the government to adopt minimization procedures requiring prompt destruction of records it determined were not foreign intelligence information. [1] They also had to direct it to destroy all records produced under the order as those procedures prescribed. [1]

Deleting records is harder than it sounds once data has moved into several systems. The oversight board recorded an example. In September 2018, NSA’s Inspector General identified a small number of data objects derived from these records that should have been deleted but were not. [2] The board attributed this to a mistaken assumption about the technical configuration of a single repository, and said the data was deleted by October 15, 2018. [2] In our reading, a deletion duty is only as good as the inventory of places the data reached.

What a recipient would need

Read as a data engineering problem, the board’s account suggests what a recipient of others’ records needs. We set these out as our own analysis; the law required none of them in these words.

  • Lineage per record: which request, which order and which source system produced it. NSA had the order code. [2]
  • Content checks as well as format checks: tests that compare a record with what is independently known, such as reconciliation with another source.
  • A defect-scoping duty: an agreement that the holder will identify every record affected by a known error, within a set time.
  • Isolation: the ability to remove one source’s records, or one period’s, without discarding everything.
  • Propagation tracking: a record of which downstream requests, decisions and reports relied on each input.

The last item is the one the program lacked in usable form. NSA could revalidate its reports and confirm they did not rely on inaccurate records. [2] It could not separate the inaccurate records from the rest of the holdings. [2]

The disclosure: what others could learn about the requests

Reports by recipients

The fourth part is disclosure. Section 603 let a person subject to a nondisclosure requirement publish aggregate counts of orders, directives or national security letters it received. [1] The law gave several structures. One was a semiannual report with national security letters in bands of 1000 starting with 0-999. [1] Another was an annual report of all national security process combined, in bands of 100 starting with 0-99. [1]

The structures differed in granularity and in time covered. A report on national security letters could cover the previous 180 days. [1] A report on orders under the surveillance statute had a lag built in. It covered a 180-day period ending, in the law’s words, “not less than 180 days prior to the date of the publication” of the report. [1] The choice of structure set how much a company could say and how current it could be.

Reports by the government

The law also expanded official reporting. The Director of the Administrative Office of the United States Courts had to report each year the number of applications, orders granted, modified and denied, and amicus appointments. [1] The Justice Department’s annual reports to Congress show what that reporting looked like later. Its report for calendar year 2020, dated April 30, 2021, counted 28 final applications for business records orders filed with the surveillance court that year. [3] It said the court did not deny any of them in whole or in part. [3]

The same report counted 6,670 national security letter requests in 2020 for information on United States persons, excluding requests for subscriber information only. [3] It said these sought information on 1,915 different United States persons. [3] It listed the letter authorities covered, including the Right to Financial Privacy Act and the two Fair Credit Reporting Act provisions. [3]

Inspector General review

Section 108 extended the Inspector General reviews of business records orders that a 2005 law had required, adding calendar years 2012 through 2014. [1] For those years it called for an examination of the minimization procedures used for such orders. [1] The examination was to cover whether those procedures “adequately protect the constitutional rights of United States persons.” [1] In our reading, Congress chose to look back at how the old authority handled data while it rebuilt the new one.

Court opinions and amici

Section 402 required a declassification review of each surveillance court decision that included a significant construction or interpretation of law. [1] It named “any novel or significant construction or interpretation of the term ‘specific selection term’” as one such case. [1] The Director of National Intelligence could meet the duty by publishing a redacted form, or could waive it on national security grounds. [1]

Section 401 required the court’s presiding judges to designate not fewer than five individuals eligible to serve as amicus curiae. [1] The court had to appoint one where an application presented a novel or significant interpretation of law, unless it found appointment not appropriate. [1] It could also appoint an amicus to provide technical expertise. [1] In our reading, this mattered for the data part. Questions such as which fields count as session-identifying information are technical questions with legal consequences.

What disclosure did and did not cover

The disclosure part made the volume and legal basis of requests more visible. It did not, by its terms, require anyone to report the accuracy of the data produced. [1] The data-integrity problems became public through the oversight board’s 2020 report and the government’s notices to the court that it described. [2] In our reading, the law’s transparency design measured how often the government asked. It did not measure whether the answers were right.

How it reached different organizations

Federal agencies using others’ data

For an agency, the program is a case study in dependence on external records. Agencies use bank data for benefits eligibility, employer data for wage verification and carrier data for identity checks. Each arrangement has the same shape: a query, a holder, a record and a report. The 2015 law shows what happens when the record part is left to the holder’s own systems and the recipient checks only format. [2]

An agency in that position would want three things from the arrangement: a defined format, a defect-reporting duty and the ability to quarantine one source. NSA had the first. [2] The oversight board’s account shows the cost of lacking the other two in workable form. [2]

Carriers and other providers

For a carrier, the law created a production obligation for records built for other uses. [1] The holder had to deliver records in a useful form, protect secrecy and minimize disruption to service. [1] It was compensated for reasonable expenses and protected from suits. [1] The oversight board’s account suggests what else the role required: the ability to find and scope defects in a production system when the recipient asks. [2]

Banks and credit unions

For banks and credit unions, the law’s main effect was on the shape of national security letters under the Right to Financial Privacy Act. [1] A letter had to name a customer, entity or account. [1] Recipients gained notice of judicial review and a process for periodic review of nondisclosure. [1] The bands in Section 603 were open to any person subject to a nondisclosure requirement, which in our reading includes a financial institution that receives a letter. [1]

Whatever the volume at a given institution, the same file discipline applies: date received, scope named, records produced, secrecy status and any termination notice.

Consumer reporting agencies

The Fair Credit Reporting Act amendments reached consumer reporting agencies directly. [1] Requests for consumer identifying information and orders for consumer reports now had to name a consumer or account. [1] For those firms, the law joined a data-accuracy regime that already governs their ordinary business. In our reading, they were better placed than carriers to answer accuracy questions, because accuracy was already part of their legal obligations to consumers.

Health systems

Health systems were not named in the law’s financial and telephone provisions. The business records authority itself applied to “tangible things” generally, and the selection-term requirement applied to any such order. [1] The more transferable point for a health system is the data one. Health systems both supply records to others and depend on records from others: claims, eligibility, lab results and referrals. The program shows what an error in a supplied record does once it drives the next request.

Objections, and where the argument stops

“It was an intelligence program, not a data program”

The first objection is that the program’s end was a policy decision about intelligence value. The board does record that NSA weighed intelligence value and costs as well as data concerns. [2] It found the records were cited in 15 intelligence reports over the program’s four-year operation. [2] It said the FBI received unique information from two of those reports. [2]

The argument does not depend on data problems being the only cause. The board said the problems, taken together, “contributed to” the decisions to delete the data in 2018 and again in 2019 and to suspend the program. [2] We claim no more than that. In our reading, a program of modest value with unscoped data defects is harder to defend than one with either problem alone.

“The errors were a tiny share”

The second objection is scale. The board’s footnote says the records from the production-system incident it described were a fraction of one percent of the overall collection. [2] The board also reported that in 2018 the government obtained 14 orders and collected more than 434 million records, including an unknown number of duplicates, involving 19 million phone numbers. [2]

The share was small, but the deletion decision turned on something else. NSA could not identify which records were affected. [2] A small error rate in an unknown location contaminates the whole set for any use that requires accuracy. That holds for a court application, a credit decision or a clinical record alike.

“The law could not have specified data quality”

The third objection is that statutes do not set data standards. That is fair. The law left format and procedure to orders and to the government. [1] It did require the government to adopt minimization procedures and to destroy records under them. [1] It did not require an accuracy standard for what providers produced. [1]

Our point is narrower than a criticism of the drafters. Any arrangement that moves data responsibility to a holder needs an accuracy and defect-scoping duty somewhere, whether in law, order or contract. In this case, the board’s account does not show one that was strong enough to let NSA keep the clean records. [2]

Where the argument stops

The public version of the board’s report contains redactions, including in the passages on data integrity. [2] We do not know the full technical causes of each defect. We do not claim that any particular provider’s systems were deficient beyond what the report says. We also do not claim that the old bulk program had better data; the board’s report is about the program under the 2015 law. [2]

Checks for leaders who build on others’ records

These are our own suggestions, drawn from the program’s history. None is a legal requirement unless an organization’s own rules make it one.

For agencies and firms that receive data

  • List every external source that feeds a decision, and for each, who can confirm a record is accurate.
  • Ask whether your checks test content or only format. Add at least one reconciliation against an independent source where the stakes justify it.
  • Write a defect-scoping duty into each data agreement: the supplier identifies every affected record, by a set deadline.
  • Keep each source’s data separable, so one source or period can be withdrawn without a full purge.
  • Record which downstream actions relied on which inputs, so a defect can be traced forward.

For firms that hold records others request

  • Keep a register of national security letters and orders: date, scope, records produced, secrecy status and any termination notice.
  • Check that each letter names a customer, entity or account where the amended statutes require it. [1]
  • Document how your production format maps to your source fields, and who can explain each field.
  • Plan how you would find every record affected by a known production defect.

For boards and audit committees

Ask management one question about every important external data feed. If the supplier told us tomorrow that some records were wrong, could we find which ones, and what we did with them? The 2015 program’s history shows the cost of the answer “no.” [2]

With hindsight

The law’s call record program started later in 2015. The oversight board said NSA represented that, since the start of the program in November 2015, it never received prohibited categories of information from providers. [2] It also found no evidence that NSA received names, addresses, financial information, cell-site location or GPS data. [2]

The data problems began to surface in 2016, according to the board. [2] NSA issued about a dozen notices to the surveillance court about compliance incidents and data-integrity problems from 2016 on. [2] On April 11, 2018, the government notified the court of inaccurate provider data and the four applications that had relied on it. [2] By May 2018, NSA had concluded it could not separate the affected records and began deleting all records produced under the law. [2] The board cited an NSA press release of June 28, 2018, titled “NSA Reports Data Deletion.” [2]

The board said NSA’s Inspector General found in September 2018 that some data objects derived from these records should have been deleted but were not. [2] It said NSA deleted that data by October 15, 2018. [2] NSA allowed its last order under the law to expire in early 2019, and had not requested records from providers since. [2] NSA told the board it would dismantle the System 1 architecture and reallocate remaining funds. [2] The board said NSA later deleted the data collected under the law. [2] It added a qualifier: whenever NSA deleted these records, it kept underlying data that had been used in disseminated intelligence reporting. [2]

The board published its report in February 2020. [2] It found the program constitutional under settled Supreme Court precedent and the two-hop collection statutorily authorized. [2] It found no abuse and judged the compliance incidents inadvertent, not willful. [2] It also found that the records were cited in 15 intelligence reports over the program’s four-year operation. [2] At that time, it said the authority was scheduled to sunset on March 15, 2020. [2] The 2015 law itself had written December 15, 2019 into the sunset provisions. [1]

The sunset took effect. The Justice Department’s annual report for 2020 said the sunset of Section 215 became effective on March 15, 2020. [3] It said the business records provision then reverted to the language and authorities originally enacted in 1998, except for investigations that began, or offenses that began or occurred, before the sunset. [3] The government still filed 28 final applications for business records orders in 2020, and the court denied none. [3]

What held up from 2015 is the shape of the national security letter changes. The 2020 report still listed the letter authorities the 2015 law amended, including the Right to Financial Privacy Act and the Fair Credit Reporting Act provisions. [3] It reported 6,670 requests that year for information on United States persons, excluding subscriber-only requests. [3] The 2015 selection-term rules also shaped the business records count. The report said thirteen final applications that year did not specifically identify an individual, account or personal device as the selection term. [3] It explained that the reporting definition is narrower than the definition for obtaining an order, so a request whose selection term was an address need not be included in that statistic. [3] As checked on November 16, 2026, that report remains on the Justice Department’s site.

What did not hold up is the idea that leaving data with its holder would make a records program simpler. The 2015 design ended bulk government holdings, as intended. [1] [2] It then depended on records whose accuracy the government could not check and the providers could not fully scope. [2] The board’s account of that dependence is the part of the story with the widest use outside intelligence work.

For agencies, banks and health systems, the same arrangement appears whenever a decision rests on another party’s records. The program’s history shows that format checks, compensation and legal protection for the holder were in place. [1] [2] A way to find and isolate wrong records was the piece the board’s account shows was missing. [2]

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

  • USA FREEDOM Act
  • FISA
  • national security letters
  • data integrity
  • call detail records
  • PCLOB