When the sender sorts the archive: records classification after the State Department email disclosure

The March 2015 disclosure that a former Secretary of State used a personal account for official email exposed one weakness above others: the person who created the messages decided which were records. NARA had already described role-based and automated alternatives.

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 March 2015 disclosure that a former Secretary of State used a personal account for official email exposed one weakness above others: the person who created the messages decided which were records. NARA had already described role-based and automated alternatives.

  1. The week On Monday, March 2, 2015, The New York Times reported that Hillary Clinton had used a personal email account, rather than a government account, for official business.
  2. Capture: getting the record into agency custody A record that never reaches an agency system depends entirely on the person who holds it.
  3. Classification: who decides what is a record This is the stage where the 2015 case is most instructive, and where data and AI practice now matters most.
  4. Retention: keeping and disposing by authority Retention determines how long a record is kept and who may authorize its destruction.
  5. Retrieval: finding the record when asked A record that is captured, classified and retained still has to be found.
  6. Objections, and where the argument is weakest One objection is that the case was about an individual’s choices, and better rules would not have changed them.
  7. How the four stages differ for state and local agencies The Federal Records Act and NARA rules apply to federal agencies.
  8. With hindsight The inspector general’s report, ESP-16-03, was issued in May 2016.

The week

On Monday, March 2, 2015, The New York Times reported that Hillary Clinton had used a personal email account, rather than a government account, for official business. The practice covered her four years as Secretary of State, from January 2009 to February 2013. The State Department’s Office of Inspector General later confirmed the core facts. It said she used an account maintained on a private server located in her New York residence, mostly through mobile devices, and did not use a department email account (State OIG, ESP-16-03).

This article is about records management and data governance. It does not address the separate investigations into the handling of classified information, or the political arguments that followed. Everything below is drawn from the public record, principally the inspector general’s May 2016 report, which reviewed the email practices of five Secretaries of State (State OIG, ESP-16-03).

The disclosure had a prehistory. According to the inspector general, State Department staff discussed in 2011 and again in 2013 whether the department needed to search personal accounts for federal records. In October and November 2014, citing its duties under the Federal Records Act, the department wrote to representatives of former Secretaries Albright, Powell, Rice and Clinton. It asked them to provide copies of any federal records in their possession, such as emails on personal accounts (State OIG, ESP-16-03).

Secretary Albright’s representative said she had not used email for department business. Secretary Rice’s representative said she had not used a personal account for official business. Representatives for Secretaries Powell and Clinton acknowledged that they had. Secretary Powell’s representative said in 2015 that he had not kept those emails (State OIG, ESP-16-03).

In December 2014, Secretary Clinton’s representative produced about 55,000 hard-copy pages, representing about 30,000 emails that she believed related to official business. The accompanying letter said it had been her practice to email department officials at their government addresses, so the department would already hold those messages in its systems (State OIG, ESP-16-03). At a news conference on March 10, 2015, she said the messages she had not provided were personal.

The National Archives and Records Administration acted quickly once the practice became public. In a letter dated March 3, 2015, the Chief Records Officer for the U.S. Government asked the State Department for a report on the potential alienation of federal email records and on what it was doing to recover them. Later in March, the department asked four of Secretary Clinton’s immediate staff to provide records from their personal accounts. They produced material during the summer of 2015 (State OIG, ESP-16-03).

The legal frame had changed only months earlier. The Presidential and Federal Records Act Amendments of 2014, signed on November 26, 2014, confirmed that federal records include electronic records. They also added a new rule. An executive agency employee may not create or send a record from a non-official electronic messaging account unless certain steps are taken. The employee must copy an official account at the time, or forward a complete copy to one within 20 days (Pub. L. 113-187)(44 U.S.C. 2911). That rule did not exist during Secretary Clinton’s tenure. The duty to preserve federal records, including email, did. The inspector general traced it back through department policy to at least 1995 (State OIG, ESP-16-03).

The March 2015 disclosure is usually discussed as a story about one person. For a records officer, a chief data officer or an agency general counsel, its more durable content is structural. The inspector general put it in those terms: weaknesses in electronic records management in the Office of the Secretary went well beyond the tenure of any one Secretary (State OIG, ESP-16-03).

Our argument is that the most important weakness was in classification. The person who created the messages also decided which of them were federal records, after the fact and alone, with no independent check. Capture, retention and retrieval all depended on that one decision. The National Archives had already published guidance and a plan on this. They described how agencies could take that decision away from individual users, through role-based rules and, eventually, machine classification (NARA Bulletin 2013-02)(NARA, Automated Electronic Records Management Report/Plan).

The question matters to agencies today for a practical reason. The volume and variety of what officials create has grown far faster than records staff. Email has been joined by chat, text, video meetings and AI-generated content. If each item still depends on its creator’s judgment, the weakness the inspector general described in 2016 is repeated across every new channel. A design that removes that dependence is the only one that scales.

We follow the four stages that any records program has to get right. They are sequential and together cover the life of a record:

  • Capture: whether the record enters a system the agency controls.
  • Classification: who decides whether a message is a record, and how.
  • Retention: how long the record is kept and how its disposal is authorized.
  • Retrieval: whether the agency can find and produce it when asked.

We then take the strongest objections, and how the same stages work for a state or local agency, whose records laws differ from the federal ones.

Capture: getting the record into agency custody

A record that never reaches an agency system depends entirely on the person who holds it. The State Department’s policy since 1995 was that email records had to be preserved in the department’s custody, generally by printing and filing them. The department introduced an electronic tool called SMART in 2009 that let staff designate emails as records. According to the inspector general, the Office of the Secretary chose not to use it for preserving emails (State OIG, ESP-16-03).

Federal regulations addressed the outside-system case directly. Since 2009, NARA’s rules had addressed agencies that let employees use email systems the agency does not operate. Those agencies had to ensure that federal records on such systems were preserved in an agency recordkeeping system (State OIG, ESP-16-03). The 2014 amendments turned that agency duty into a personal one, with the 20-day forwarding rule and possible disciplinary action for intentional violations (44 U.S.C. 2911).

Why copying a colleague is not capture

The December 2014 letter argued that messages sent to officials’ government addresses were already in the department’s systems. The inspector general rejected that as a method of preservation. Sending email from a personal account to colleagues’ department accounts was not an appropriate way to preserve federal records (State OIG, ESP-16-03).

The reasoning matters beyond this case. A message in a recipient’s mailbox is not managed as the sender’s record. It may be deleted under the recipient’s schedule, filed in the recipient’s context, or never identified at all. Messages to people outside the department are not captured anywhere. The inspector general found that the December 2014 production did not include some messages from an external contact about department business (State OIG, ESP-16-03).

Capture by design

The durable fix is to make the agency’s own systems the only practical path for official business. That means government accounts for every official, including the most senior, configured to work on the devices those officials actually use. It means journaling or archiving at the mail server, so capture does not depend on anyone filing anything. And it means blocking or monitoring automatic forwarding to personal accounts, which creates the reverse leak.

The 2014 amendments made the copy-or-forward rule the legal floor for exceptions (44 U.S.C. 2911). An agency can make it easier to follow. Official accounts that work well on mobile devices remove the most common reason officials give for using personal ones. A simple, approved way to forward a message from a personal account into the official one, with a confirmation, turns a legal duty into a routine.

The device is where capture succeeds or fails

The inspector general said the personal account was used predominantly through mobile devices (State OIG, ESP-16-03). That detail matters more than it first appears. Senior officials live on their phones. If the official account is awkward on the device they carry, or requires a second device, they will drift toward whatever works. Capture policy that ignores the device is policy written for a desk that senior officials rarely sit at.

Agencies have more options in 2026 than in 2009. Managed mobile devices can run official email and messaging in a separated work profile, with archiving at the server and no dependence on the user. The trade-offs are real: security requirements for senior officials’ devices are strict, some secure configurations are inconvenient, and officials often want one device for everything. A records officer and a chief information security officer who design the senior official’s device together will produce something both usable and capturable. Designing it separately tends to produce one or the other.

Messages from outside the agency

Capture also has to account for correspondents who are not agency employees. Ambassadors, foreign officials, contractors, academics and members of the public write to senior officials. Those messages can be records. They reach the agency only if they arrive at an official address. When an official corresponds from a personal account, outside parties will reply there, and nothing on the agency side sees the exchange. Publishing official addresses, and redirecting correspondents who use personal ones, is part of capture.

Evidence that capture works

An auditor assessing capture would look for a few artifacts. A list of all senior officials and confirmation that each has an official account in active use. Archiving settings showing that all mail through those accounts is retained at the server. Records of training on the 20-day rule. A log of messages forwarded from personal accounts, which should be small but not zero. An agency that reports zero forwarded messages from a large senior staff should ask whether the rule is understood.

Classification: who decides what is a record

This is the stage where the 2015 case is most instructive, and where data and AI practice now matters most. A federal record is defined by its content and function. It is recorded information made or received in connection with agency business. It must also be appropriate for preservation as evidence of the agency’s organization, functions, policies, decisions, procedures, operations or other activities. The definition does not depend on the account or device. Since the 2014 amendments, the Archivist’s determination of whether something is a record is binding on agencies (Pub. L. 113-187).

In the December 2014 production, the classification was done by the account holder’s representatives, after the fact, and without review by anyone at the department. That is the pattern NARA had described a few months earlier as the core weakness of federal email management, in agencies generally as well as at State.

A binding referee for hard cases

The 2014 amendments gave classification a referee. The Archivist’s determination of whether recorded information is a federal record is binding on agencies (Pub. L. 113-187). That matters when an agency and an official disagree, or when a new kind of content does not fit existing guidance. An agency facing a genuinely hard question, such as whether a category of messages or AI outputs is a record, can document its reasoning and, where the stakes justify it, seek NARA’s view. A documented decision with NARA input is far easier to defend later than one made alone by the person whose conduct is at issue.

NARA had already named the problem

In September 2014, NARA published a report and plan on automating electronic records management under the Managing Government Records Directive. It said most agencies relied on individual staff to capture and categorize their electronic records. It said end users found it burdensome to make a separate decision about each item. Relying on busy staff led to inconsistent capture, and manual processes would not scale to the volume of email (NARA, Automated Electronic Records Management Report/Plan).

The report laid out a range of approaches. Manual management sat at one end, and NARA judged it very difficult to make consistent. Rule-based automation came next, including role-based rules. Then workflow capture, modular tools and, at the far end, autocategorization using machine learning (NARA, Automated Electronic Records Management Report/Plan). Read in March 2015, it described almost exactly the failure the disclosure revealed, and the alternatives.

Classification by role: Capstone

NARA’s August 2013 Capstone bulletin offered the simplest alternative. Under Capstone, email in the accounts of officials at or near the top of an agency is captured and scheduled as permanent, by account rather than by message. Other accounts can be scheduled as temporary and kept for a set period (NARA Bulletin 2013-02). The individual no longer decides which messages are records. The role decides.

Capstone moves the classification decision from the end of the process to the beginning. Instead of thousands of message-level judgments made by the user, it requires one designation per account, made by the records office, documented and reviewable. The trade-off is over-retention. Personal and trivial messages in a Capstone account are kept along with the records. NARA accepted that trade-off because the alternative was under-retention of the most important records an agency creates.

Capstone also has limits. It only works for messages in official accounts. A senior official who used a personal account would leave the Capstone net entirely. Capture and classification depend on each other, which is why they have to be designed together.

Classification by content: machine learning

NARA’s 2014 report described autocategorization as the most advanced approach. An expert trains software on a sample of categorized records, and the software learns to assign new items to retention categories by their content. The report noted that this could classify email by subject or function rather than by account, which Capstone cannot do. It also warned that the technology was still developing and that training required significant expert time. Schedules might need consolidating into fewer, clearly separated categories for training to work (NARA, Automated Electronic Records Management Report/Plan).

For a chief data officer, those cautions read like a modern machine-learning governance checklist. The training set has to be representative. The categories have to be mutually exclusive. Accuracy has to be measured against human decisions, and the error rate has to be acceptable for the consequence of a mistake. Destroying a permanent record by misclassification is a worse error than keeping a temporary one too long. A classifier for records should be tuned accordingly, and its thresholds documented.

Measuring a classifier

Any automated classification, whether by rule or by model, needs measurement that a records officer can explain to NARA, an inspector general or a court. Two numbers matter most. The first is how many true records the system misses, which drives the risk of loss. The second is how many non-records it keeps, which drives cost and exposure in litigation and records requests. A system tuned to miss nothing will keep a lot; a system tuned to keep little will miss some.

Measurement comes from sampling. Records staff review a random sample of items the system classified, in each category, at a set interval. They record agreement and disagreement, and the disagreements become new training data or rule changes. The sample results, the thresholds in use and the reasons for them form the evidence file. Without that file, an agency cannot show that its automated decisions are better than the manual ones they replaced. NARA’s report described the aim as training until the software approaches the decisions a human would make (NARA, Automated Electronic Records Management Report/Plan).

Who owns the rules

Classification rules sit across several owners. The records officer owns the schedules and the decision whether something is a record. The chief data officer, where an agency has one, owns data governance and often the platforms where classification runs. The chief information officer runs the email and messaging systems. General counsel advises on what is a record in hard cases and on litigation exposure. When those roles do not meet, automated classification tends to be built by whoever controls the platform, without the records officer’s schedule as its target.

A simple governance structure fixes most of this. One documented owner for each classification rule or model, a change process that requires the records officer’s approval, and a periodic joint review of measurement results. The same structure serves for AI-generated content, where the questions are new but the owners are the same.

Classification of new kinds of content

The classification problem has grown since 2015. Agencies now generate text messages, chat threads, collaboration posts, recorded meetings and, increasingly, summaries and drafts produced by generative AI assistants. Each raises the same question the 2015 case raised: is this a record, and who decides? Leaving the decision to the person who created the content repeats the weakness the inspector general described.

The approaches NARA described in 2014 still apply. Role-based rules can cover chat and messages in senior officials’ accounts. Workflow capture can cover records generated inside case systems. Content-based classification can handle the rest, with the safeguards above. What changes with generative AI is that the system, not a person, may create the content. A meeting summary produced automatically may be the only written account of a decision. Agencies need to decide in advance whether such outputs are records, rather than discover the question in litigation.

Retention: keeping and disposing by authority

Retention determines how long a record is kept and who may authorize its destruction. Federal records may be destroyed only under a schedule approved by the Archivist. Records judged permanent are eventually transferred to the National Archives. Records that are removed, destroyed or lost without authority must be reported, and the Federal Records Act requires agencies to establish safeguards against removal or loss (State OIG, ESP-16-03).

NARA’s March 3, 2015 letter used the language of that framework. It asked about the potential alienation of federal records, meaning records taken out of agency control, and about steps to recover them (State OIG, ESP-16-03). Later, NARA asked the department to contact the email providers associated with the personal accounts of Secretaries Powell and Clinton to see whether any messages remained on their servers (State OIG, ESP-16-03).

Personal messages and the deletion decision

Officials are entitled to personal correspondence, and agencies do not keep it as records. The question for retention is who sorts the personal from the official before any deletion happens. In a captured system, a records office or a rule makes that sort, and only temporary records reach their disposal date. Outside a captured system, the account holder sorts and deletes, and the agency cannot verify what was destroyed.

The inspector general found gaps in the production that illustrate the risk. It contained no email from the first weeks of the tenure, January 21 to March 17, 2009 for received messages and to April 12, 2009 for sent messages. The inspector general also found official messages in that period through other channels, including 19 emails with a Defense Department official that were not in the production (State OIG, ESP-16-03). The record could not be reconstructed from the account holder’s side.

Departure as a retention checkpoint

Departure is the last moment an agency can easily recover records from an individual. The inspector general recommended that all departing officials in the Office of the Secretary, including the Secretary, sign a separation form certifying that they had surrendered all federal records. It also recommended that incoming officials be briefed on their records duties, including for personal accounts (State OIG, ESP-16-03).

Those recommendations translate into a standard exit control. The records officer is notified of every senior departure in advance. Holdings in official accounts and devices are confirmed before access is removed. The departing official signs a certification covering personal accounts and devices. Exceptions go to counsel. The control is cheap. Its absence is what makes later recovery so expensive.

When records leave custody

The 2015 case also showed what an agency must do when records have already left its custody. NARA’s March 3 letter framed it as a reporting and recovery duty (State OIG, ESP-16-03). The agency identifies what may be missing, reports to NARA, and tries to recover it, from the individual, from the individual’s representatives and, where possible, from service providers. NARA’s later request that the department contact the email providers shows how far recovery may need to reach (State OIG, ESP-16-03).

Recovery is slow and often incomplete. Commercial providers have their own retention practices, and messages deleted years earlier may be gone. The inspector general reported that, as of May 2016, the department had not received a response on its request concerning Secretary Powell’s account (State OIG, ESP-16-03). The practical lesson for an agency is that recovery is the most expensive and least reliable stage of records management. Every control earlier in the chain exists to avoid needing it.

Holds override schedules

Retention has an exception that every records program must handle: legal holds. When litigation is reasonably anticipated, or an investigation or records request is pending, records that would otherwise reach their disposal date must be preserved. A hold overrides the schedule. Applying a hold depends on knowing where the relevant records are, which brings the analysis back to capture. Records in a personal account cannot be held by the agency at all. It can only ask the account holder to preserve them.

Automated retention makes holds both easier and more important. Easier, because a hold can be applied to every account or item matching a criterion in one step. More important, because automated disposal that ignores a hold destroys records at machine speed. The hold process needs a direct line into whatever system executes disposal, and a test that confirms held items are actually exempt.

Schedules that machines can execute

Retention schedules were historically written for people. Many describe records by subject in terms a human can apply but a system cannot. NARA’s 2014 report noted that rule-based automation may require rescheduling, so that schedule items can be expressed as executable rules on metadata, roles or other features (NARA, Automated Electronic Records Management Report/Plan). Capstone is itself an example: a schedule keyed to an account.

For an agency planning automated classification, the schedule is the training target. Big-bucket schedules with fewer, broader categories are easier to automate and easier to apply consistently. The cost is less precision in how long individual records are kept. That is usually a good trade for email and messages, where precision was never achieved by hand anyway.

Retrieval: finding the record when asked

A record that is captured, classified and retained still has to be found. Agencies receive Freedom of Information Act requests, congressional requests, litigation holds and inspector general reviews. Each requires searching the right places and producing complete results. The inspector general’s report described a limited ability to retrieve email records and inaccessibility of electronic files as weaknesses that contributed to lost records across the department (State OIG, ESP-16-03).

Messages in a personal account are outside every search an agency can run. A request for a senior official’s correspondence will return messages in that official’s government account and in colleagues’ accounts. It will miss what stayed in the personal account and what went to outside parties. The agency’s answer is incomplete, and it may not know that it is.

Paper as a retrieval barrier

The December 2014 production was paper, about 55,000 pages (State OIG, ESP-16-03). The department’s policy at the time still treated print-and-file as the default way to preserve email records. Paper satisfies preservation. It does very little for retrieval. Searching, deduplicating, redacting and producing paper copies of email is slow, and each step introduces error. The department later reviewed and released the produced emails in stages through 2015 and early 2016, in response to Freedom of Information Act litigation.

Electronic capture with metadata makes retrieval a query rather than a project. NARA’s Capstone bulletin required agencies to keep the metadata that makes retrieval possible, such as dates and sender and recipient names and addresses. Email records also had to remain accessible to appropriate staff for as long as needed to conduct agency business (NARA Bulletin 2013-02).

One search across many channels

By 2026, a request for a senior official’s communications may need to search email, chat, text messages, collaboration platforms, recorded meetings and AI-generated summaries. Each channel may sit in a different system with a different search tool and a different retention setting. Retrieval is only as complete as the channel with the weakest capture. Agencies that consolidated their search into a single compliance or e-discovery platform covering all official channels can answer such requests with one query. Agencies that did not must coordinate several searches and reconcile the results.

Retrieval as a measurable service

Retrieval can be measured. An agency can test it by running sample searches for a senior official’s correspondence and comparing results across systems. It can track how long requests take, how often a search had to be repeated, and how often responsive material later surfaced that the search had missed. Those numbers show whether capture and classification are working, because retrieval failures usually trace back to them.

Proactive disclosure is the other side of retrieval. NARA’s 2014 report noted that well-designed automation would serve purposes beyond records management, including privacy, security, e-discovery, Freedom of Information Act processing and proactive disclosure of government information (NARA, Automated Electronic Records Management Report/Plan). An agency that can classify and find its records reliably can also publish the ones the public is entitled to see without waiting for a request. An agency that cannot find them is left responding to requests one at a time, usually late.

Retrieval also connects records management to data governance more broadly. The same index that answers a records request supports e-discovery, security investigations and, increasingly, the data pipelines that feed analytics and AI tools. An agency that cannot find its senior officials’ email for a records request cannot safely use that email for anything else either.

Objections, and where the argument is weakest

One objection is that the case was about an individual’s choices, and better rules would not have changed them. There is truth in that. A determined official can work around most controls. The inspector general also found that the department rarely enforced its rules against senior officials. Apart from one ambassador, it could find no case in which the department had begun disciplinary action against a senior official for using non-department systems for day-to-day work (State OIG, ESP-16-03). Our argument is that design changes the default. When the official system is the easy path and capture is automatic, avoiding it becomes a visible choice rather than a habit.

A second objection is that role-based capture keeps too much. Capstone accounts hold personal messages, routine notices and drafts alongside records. That raises storage costs and, more seriously, expands what can be requested in litigation and under the Freedom of Information Act. The objection is fair. NARA accepted over-retention for senior accounts because the cost of under-retention there is higher, and agencies can still filter clearly non-record items with rules.

A third objection is that machine classification is not ready for records decisions. In 2014, NARA said as much about the state of the technology (NARA, Automated Electronic Records Management Report/Plan). Content-based classification has improved since, but the objection still has force where a misclassification could destroy a permanent record. We would apply machine classification first to retention of temporary records and to triage, with human review for anything proposed for destruction from senior accounts.

A fourth objection concerns officials’ privacy. Role-based capture of a senior official’s account will sweep in personal messages, and broad search tools will surface them in records requests and investigations. Officials may reasonably resist. The answer is not to stop capture but to keep personal business off official accounts by policy, and to give records and legal staff clear rules for withholding personal content that does get captured. The approach mirrors the 20-day rule: official business belongs in official accounts, and personal business belongs elsewhere.

The argument is weakest at the very top of an organization. Senior officials have the most discretion, the most demands on their time and the most influence over the people who would enforce controls on them. Technology can make compliance easy. It cannot make it mandatory for someone who chooses otherwise. Leadership commitment and the records officer’s standing in the agency remain the deciding factors.

How the four stages differ for state and local agencies

The Federal Records Act and NARA rules apply to federal agencies. State and local agencies follow their own public records and retention laws, which vary widely. Most states have a state archives or records commission that issues retention schedules for state agencies and for local governments. Most also have public records acts that define what is a public record, often by content and function in terms similar to the federal definition.

Capture

Use of personal accounts and devices by state and local officials has been a recurring issue under state public records laws. Courts in several states have treated official business conducted on personal accounts as subject to those laws. A county or city should assume that official messages on personal devices can be requested, and design capture accordingly. Providing official accounts and mobile access to elected officials and board members, who often have none, is the first step.

Classification and retention

State general retention schedules often classify email by content, with different periods for correspondence, transitory messages and policy records. That puts the classification decision back on individuals, with the weakness described above. A state or local agency that wants a role-based approach like Capstone should first check what its state archives permits, since a schedule must authorize any disposal.

Retrieval

Smaller jurisdictions often lack e-discovery tools and rely on staff to search their own mailboxes in response to requests. That repeats the classification weakness at the retrieval stage: the person whose records are requested decides what is responsive. Centralized search, even through the basic compliance features many email platforms include, moves that decision to a records or legal officer.

Large and small federal agencies

Even among federal agencies the stages play out differently. A large department has a records office, an e-discovery team and enterprise archiving, and its main risk is the gap between policy and the behavior of senior officials. A small commission or board may have one records officer with other duties, a cloud email service with default retention settings, and no dedicated search tools. For the small agency, the most valuable step is usually turning on and configuring the archiving and retention features already included in its email platform, and designating its few senior accounts under Capstone. The large agency’s most valuable step is enforcement at the top.

What carries across

Across all levels of government, the four stages are the same, and so is the central question. Whoever decides whether a message is a record controls everything that follows. The design goal is to make that decision once, at the right level, with a rule or a reviewable process, rather than many times, alone, by the person whose conduct the record documents.

With hindsight

The inspector general’s report, ESP-16-03, was issued in May 2016. It reviewed the tenures of Secretaries Albright, Powell, Rice, Clinton and Kerry. It concluded that email practices and compliance varied across them, and that the Office of the Secretary had been slow to recognize and manage the legal requirements and cybersecurity risks of electronic communications. It made eight recommendations. They covered guidance on personal accounts, an end to print-and-file in favor of electronic systems, and inventories of the Office of the Secretary’s files. Others addressed departing officials, separation certifications, a quality assurance plan and security measures (State OIG, ESP-16-03).

The department concurred. Its Bureau of Administration said it aimed to end print-and-file as department policy for retaining email by December 31, 2016, the date by which an August 2012 OMB and NARA memorandum required agencies to manage all email records electronically (State OIG, ESP-16-03). The report also noted that Secretary Kerry’s emails were being retained under the Capstone approach (State OIG, ESP-16-03).

Congress also acted on the department specifically. The Consolidated Appropriations Act of 2016, enacted on December 18, 2015, required the State Department to update its policies and directives to comply with federal records requirements. The inspector general described it that way (State OIG, ESP-16-03). The department had already revised its Foreign Affairs Manual in October 2015 to reflect the 20-day rule and to discourage personal accounts for official business (State OIG, ESP-16-03).

Across the federal government, the direction NARA set before March 2015 held. Capstone came into wide use for agency email, and NARA issued a General Records Schedule, GRS 6.1, for email managed under it. The electronic-records deadlines moved, though. The 2012 directive had set 2019 for managing permanent electronic records electronically. A 2019 memorandum, M-19-21, set new targets. A December 2022 update, M-23-07, required agencies to manage permanent and temporary records electronically by June 30, 2024. After that date, NARA would no longer accept new transfers of analog records unless it granted a limited exception (OMB and NARA, M-23-07).

The 20-day rule in the 2014 amendments became the reference point for later disputes about officials’ use of personal accounts and messaging apps. It defines electronic messages broadly, as email and other electronic messaging systems used to communicate between individuals (44 U.S.C. 2911). Later episodes involving text and encrypted messaging applications at several agencies tested the same rule under different technology.

Machine classification developed more slowly in records management than NARA’s 2014 report anticipated, and faster everywhere else. Agencies made more use of rule-based and role-based approaches than of content-based classification. By 2026, the arrival of generative AI tools in office software had made the classification question more urgent. These tools produce meeting summaries, drafts and chat responses whose record status someone has to decide.

What held up from March 2015? The structural reading did. The inspector general said the weaknesses went well beyond any one Secretary (State OIG, ESP-16-03), and the federal response was structural: rules for personal accounts, role-based capture, and electronic recordkeeping deadlines. What held up less well was the timeline. Deadlines slipped, and the hardest part, classification of content that does not fit an account-based rule, remained largely unsolved. The principle that the creator should not be the only classifier is as relevant to an AI-generated summary in 2026 as it was to a Secretary’s email in 2015.

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

  • Federal Records Act
  • NARA
  • Capstone
  • email records
  • records classification
  • State Department