Anthem’s data left through an administrator’s login, which made encryption the wrong debate

Anthem disclosed in February 2015 that attackers had taken records on up to 78.8 million people. They used valid database credentials, so the controls that mattered were privilege, retention and detection, not encryption at rest.

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

Anthem disclosed in February 2015 that attackers had taken records on up to 78.8 million people. They used valid database credentials, so the controls that mattered were privilege, retention and detection, not encryption at rest.

  1. The week On the evening of Wednesday, February 4, 2015, Anthem, Inc. announced that it had been the target of what it called a sophisticated external cyber attack.
  2. The way in: a phishing email at a subsidiary The federal investigation found that the attackers entered through spear phishing emails sent to an Anthem subsidiary.
  3. The privilege: credentials with legitimate access to the data The decisive step came after the phishing email: the attackers’ acquisition of credentials that could query the database where member records were stored.
  4. The store: a decade of records in one place The credentials opened a data warehouse that held records on up to 78.8 million people, going back to 2004 (California Department of Insurance).
  5. The exit: weeks of exfiltration before anyone noticed OCR found that the attackers stole data between December 2, 2014 and January 27, 2015.
  6. What the chain meant for different organizations Every regulated organization in 2015 had some version of each stage: a workforce exposed to phishing, administrators with standing access, warehouses accumulating history and monitoring weak on database activity.
  7. The case for encryption, and why it does not change the conclusion The strongest objection to this argument holds that encryption deserved its prominence.
  8. With hindsight Anthem’s figure of 78.8 million affected people held.

The week

On the evening of Wednesday, February 4, 2015, Anthem, Inc. announced that it had been the target of what it called a sophisticated external cyber attack. Attackers had gained unauthorized access to some of its information technology systems and obtained personal information on current and former members and employees. The data included names, birthdays, health care identification and Social Security numbers, street addresses, email addresses and employment information, including income data. Anthem said it had no evidence that credit card or medical information, such as claims, test results or diagnostic codes, had been targeted or obtained (Anthem Form 10-K for 2014).

Anthem was then one of the largest health benefits companies in the United States. The discovery had come a week earlier. According to the California Department of Insurance, a database administrator noticed on January 29 that his credentials were being used to run queries he had not run. Anthem shut down access to the database, required its employees to reset their passwords, notified the FBI and retained a cybersecurity firm (California Department of Insurance). In its annual report, filed later that month, Anthem said it had acted immediately to remediate the vulnerability and would provide credit monitoring and identity protection to those affected (Anthem 10-K).

The scale emerged over the following weeks. The California regulator reported that records of up to 78.8 million current and former enrollees might be affected, including minors. The records went back to 2004 and included people enrolled in independent Blue Cross and Blue Shield plans, as well as Anthem’s own. Anthem had about 37 million current enrollees at the time, so most of the people in the database were not current Anthem members. About 13.5 million of the records accessed related to Californians (California Department of Insurance). Anthem offered two years of free credit monitoring and identity protection to those affected.

Anthem’s structure explains why the population was so broad. It was an independent licensee of the Blue Cross and Blue Shield Association in fourteen states, and it also ran other plans and administered coverage for employers. A person could appear in its systems for several reasons. They might once have held an Anthem policy, or their employer might have used Anthem to administer a self-funded plan. They might also have belonged to another Blue plan and received care in Anthem’s territory. The breach report Anthem later filed covered data it held for its affiliated plans and for other covered-entity health plans (HHS Office for Civil Rights, 2018).

The response engaged many authorities at once. State insurance regulators opened examinations, with California a lead state in a multi-state review coordinated through the National Association of Insurance Commissioners. State attorneys general opened investigations. By the time Anthem filed its annual report on February 24, it was responding to a number of governmental inquiries and faced purported class action lawsuits (Anthem 10-K). Regulators also warned consumers about scam emails and calls that tried to exploit the news.

The incident quickly reached the policy debate in Washington. On February 10, the President’s homeland security adviser cited the Anthem breach, announced the previous week, in a speech on cyber threats, alongside earlier attacks on retailers, a bank and Sony Pictures (White House, February 10, 2015).

Much of the first week’s public commentary focused on a single question: whether the stolen data had been encrypted while stored in the database. HIPAA’s Security Rule treated encryption of stored data as an addressable specification, which a covered entity could implement or document an equivalent alternative to. The debate was understandable. It was also, as the rest of this article argues, aimed at the wrong control. Details of the attack chain came out later, through the regulator’s report, a federal indictment and enforcement settlements. They showed that the data had been taken by someone using valid credentials with legitimate access to the database.

The Anthem breach is often filed under encryption, or under nation-state espionage. Both labels miss what the attack chain shows. The data left through a database administrator’s valid credentials, after an intrusion that began with phishing at a subsidiary. Encryption at rest would not have stopped a query run with those credentials. The controls that would have mattered govern who can act as an administrator, how much data one account can reach, how long data is kept, and whether anyone notices unusual queries.

The argument follows the attack in four stages, each a separate H2 below: the way in, the privilege the attackers obtained, the store of data that privilege opened, and the long period before anyone noticed. Those four stages are mutually exclusive and together describe how the data left. Two further sections follow. One looks at what the chain meant for different kinds of organizations. The last tests the argument against the strongest case for encryption.

The way in: a phishing email at a subsidiary

The federal investigation found that the attackers entered through spear phishing emails sent to an Anthem subsidiary. At least one employee responded, which gave the attackers a foothold (HHS OCR). The Justice Department’s later indictment alleged that the group began breaking into the networks of Anthem and three other U.S. companies in February 2014, and then installed malware and tools to extend their access (Justice Department, May 9, 2019).

Why a subsidiary mattered

Large health insurers grew by acquisition, and Anthem was no exception. Acquired companies brought their own networks, email systems and user populations. Integrating them into a single security model took years. Where the subsidiary’s network was connected to the parent’s, a compromise at the edge of the group could become a route to its center.

For a security team, the subsidiary is often the least visible part of the estate. Its users may follow different training, its email filtering may differ and its accounts may carry trust into the parent’s directory. An attacker choosing where to send phishing emails will look for exactly those differences.

Why phishing could not be the main control

Phishing training and email filtering reduce the number of successful lures. They cannot reduce it to zero across tens of thousands of employees. An organization of Anthem’s size had to assume that some employee, somewhere, would eventually respond to a well-crafted message. The question was what that employee’s account could reach next.

That framing shifts investment. Effort spent on making each user perfect yields less than effort spent on limiting what any one compromised user can do. The first stage of the Anthem chain was ordinary. Every large organization was exposed to it in 2015, and still is.

From foothold to database

A foothold on one workstation is rarely where the data is. Attackers in campaigns like this one typically spend weeks or months mapping the network, collecting credentials stored or cached on machines they control and moving toward systems that hold what they want. The indictment described the group installing malware and tools to further compromise the victims’ networks before identifying data of interest (Justice Department). OCR’s director, announcing the settlement, described attackers who had gained access to Anthem’s system to harvest passwords (HHS OCR).

That middle phase is where most of an attacker’s time goes, and it is where an organization has the most chances to notice. Each move between machines uses a credential, generates a log entry somewhere and touches a system that has a normal pattern of use. The long gap between the February 2014 start and the December 2014 theft was a period in which the attackers were present but not yet taking data at scale.

What an organization would do about the entry stage

Reasonable steps at this stage were well understood in 2015, even if unevenly applied. They included filtering and sandboxing of email attachments and links, rapid reporting channels so employees could flag suspicious messages, and isolation of acquired networks until they met the parent’s security baseline. Multifactor authentication for remote access and email would have raised the cost of using stolen passwords. A determined group could defeat any one of these. Each would have reduced the number of usable footholds and slowed the movement from one to the next.

The weightier decision concerned connectivity. Leaving an acquired company’s network flat and trusted to the parent made integration easy and attack paths short. Keeping it segmented until its controls matched the parent’s cost time and created friction for staff. Most organizations chose speed. The Anthem chain shows the price of that choice.

The privilege: credentials with legitimate access to the data

The decisive step came after the phishing email: the attackers’ acquisition of credentials that could query the database where member records were stored. The California Department of Insurance reported that the intrusion was discovered when a database administrator saw his own credentials being used without his knowledge (California Department of Insurance). The attackers were operating as a trusted insider.

What credential theft defeats

Controls that protect data from outsiders assume that the outsider lacks valid access. Network firewalls, encryption of files on disk and access control lists all let an authorized account through. Once an attacker holds a database administrator’s credentials, the database will return whatever that administrator could see. Encryption at rest protects against stolen disks or backups. It does not protect against a query from an account entitled to read the decrypted data.

OCR’s findings reflected this. Among other things, it found that Anthem had failed to implement adequate minimum access controls to prevent the attackers from accessing sensitive data, beginning as early as February 2014 (HHS OCR). Access control, not encryption, was the failure named.

How privileged access is usually organized

In most large organizations in 2015, database administrators held standing privileges. Their accounts could read and export production data at any time, because their work required occasional broad access. Those accounts often authenticated with passwords alone. They were sometimes shared among team members, or used by scripts.

The alternative designs were known. Privileged access management tools could vault administrator credentials, issue them for a session, require a second factor and record what was done. Separate accounts could be used for administration and for email, so that a phished email account did not lead to database access. Administrators could be given access to manage the database without default rights to read the data in it. Each of these adds friction to legitimate work, which is why many organizations had not adopted them.

What a privileged access program involves

A practical program starts with an inventory of every account that can read or export bulk sensitive data, including service accounts and accounts used by vendors. It then ranks them by the volume of data each can reach. The highest-ranked accounts move first into a vault, with check-out, a second factor, session recording and automatic rotation of passwords.

The program also changes roles. Database administrators keep the ability to maintain systems, while bulk reads of member data move to tightly controlled service paths with their own logging. Emergency access exists, but its use triggers review. Owners for each privileged account are named, and quarterly reviews confirm whether each account still needs its rights.

The trade-offs are real. Vaulting slows routine work, and administrators resist tools that interrupt them during an outage. Service accounts are hard to bring under control because applications break when their passwords change. The programs that succeed start narrow, with the accounts that can reach the most data, and expand once the operating routine is accepted.

Service accounts and vendor access

Human administrators are the visible part of privileged access. Service accounts used by applications, extract jobs and reporting tools are often larger in number and less watched. They tend to have broad read rights, passwords that never change and no second factor, because no person logs in with them. An attacker who finds a service account’s password in a configuration file or script can use it as easily as a stolen administrator login.

Vendor access adds another layer. Outsourced database support, analytics contractors and software suppliers often hold accounts on production systems. Their staff work under the vendor’s controls, not the organization’s. A privileged access program has to cover these accounts as well, with named individuals, time-limited access and the same logging, written into the contract.

A second factor where it counts most

Multifactor authentication was common for remote access by 2015 but rare for internal administrative logins. The reasoning was that the internal network was trusted. Anthem’s chain shows why that reasoning failed: by the time the attackers used the administrator’s credentials, they were already inside. Requiring a second factor for administrative access to sensitive databases, even from inside the network, would have forced the attackers to defeat another control at the most valuable step. The cost was modest compared with a full privileged access program, which made it a sensible first move for organizations that could not do everything at once.

How an auditor or examiner would test it

Under HIPAA, a covered entity must conduct a risk analysis and implement access controls appropriate to that analysis. OCR found that Anthem had failed to conduct an enterprise-wide risk analysis (HHS OCR). An auditor testing the privilege stage would ask for evidence that bears on both requirements. That evidence would include a current list of privileged accounts with owners and review dates, and the risk analysis entries that cover the systems those accounts reach. It would also include a sample of privileged sessions with their approvals and recordings, and proof that terminated administrators lost access promptly.

An organization that can produce that evidence quickly usually has a working program. One that has to assemble it by hand, from several teams, usually does not. The time it takes to answer the request is itself a useful measure.

The store: a decade of records in one place

The credentials opened a data warehouse that held records on up to 78.8 million people, going back to 2004 (California Department of Insurance). Anthem had around 37 million current members. Most of the people whose data was taken were former members, employees or members of other Blue plans. The size of the breach was a function of what the organization had chosen to keep and combine.

Why warehouses concentrate risk

Data warehouses exist to combine data from many operational systems so that analysts can study it. Enrollment, claims, provider and customer service data flow in, are cleaned and are joined. The result is valuable for analytics, fraud detection and reporting. It is also a single location holding identity data for everyone the organization has ever covered.

Operational systems usually hold current data needed for transactions. Warehouses accumulate history. Unless someone decides to delete it, a warehouse built in 2004 still contains 2004 records in 2015. Each year of retention adds people to the potential breach population without adding much analytic value for most uses.

Why identity data, not medical data, was the target

Anthem said the attackers did not appear to have taken claims or diagnostic data. What they took was the combination of names, dates of birth, Social Security numbers, addresses, employment and income (Anthem 10-K). That set is useful for identity theft, for building profiles of individuals and for targeting further phishing. The indictment described the data sought as personally identifiable information and confidential business information (Justice Department).

For a health plan, Social Security numbers had historically been used as member identifiers or stored alongside them. Medicare itself used numbers based on Social Security numbers on its cards until it began replacing them in 2018. Many plans had moved away from using Social Security numbers on cards but kept them in back-end systems for matching and tax reporting. The warehouse inherited that practice.

People who never chose the organization

The breach population included members of other Blue plans and people whose employers had used Anthem to administer benefits. These people had no direct relationship with Anthem and no way to know their data was there. The Blue system’s arrangements let members receive care across states, with claims processed by the local licensee. That design served members well. It also meant that one licensee’s warehouse held data about members of many others.

For health systems and employers, the lesson concerned data they had passed to others. An employer sponsoring a self-funded plan had sent eligibility files to its administrator. A health system had sent claims. Each had little visibility into how long the recipient kept the data or where it was combined.

Minimization as a security control

The practical response is data minimization treated as security engineering rather than privacy policy. It means setting retention periods for warehouse data based on need, purging or archiving older records into stores with tighter access, and replacing direct identifiers with tokens where analysts do not need them. Analysts studying claims trends rarely need Social Security numbers. Matching processes that do need them can call a separate, more tightly controlled service.

These steps cost money and slow analytics teams, who prefer complete history in one place. They also conflict with legitimate retention obligations, such as records kept for regulatory or legal reasons. Those obligations can usually be met with archived, access-restricted copies rather than live warehouse tables. The design choice is between convenience for analysts and the size of the worst day.

Who owns the decision to keep data

Retention in warehouses often has no owner. The team that built the warehouse kept everything because storage was cheap and deleting data might break a report. Business units assumed the data team had a policy. Legal and compliance teams set retention schedules for records but rarely mapped them to analytic copies. The result was indefinite retention by default.

Fixing that requires a named owner for each major data store, usually a business executive rather than the technology team, who approves what is kept and for how long. It also requires a process for deletion that analytics teams trust, with notice before data is removed and a route to request exceptions. Without an owner, retention keeps growing.

The size of the store set the cost of the breach

Many of the costs of a breach scale with the number of people affected rather than with the sophistication of the attack. Notification letters, call center capacity, credit monitoring, identity restoration services and legal exposure all grow with the population. Anthem offered two years of credit monitoring and identity protection to everyone affected (California Department of Insurance). For a population in the tens of millions, that single commitment was a major operational effort.

Reducing the store reduces those costs before any incident happens. Each year of history purged, and each former member removed from active tables, lowers the worst-case notification population. That makes minimization one of the few security controls whose benefit can be estimated in advance, by counting the records it removes.

The exit: weeks of exfiltration before anyone noticed

OCR found that the attackers stole data between December 2, 2014 and January 27, 2015. The intrusion was discovered on January 29 (HHS OCR). The indictment and the multistate settlement both put the start of the intrusion in February 2014 (Justice Department) (Connecticut Attorney General, 2020). The attackers were inside for many months before they began removing data, and removed it over nearly two months before the theft was detected.

Who noticed, and how

The discovery came from a person, not a system. A database administrator recognized activity under his own name that he had not performed (California Department of Insurance). That is a good outcome for a person and a weak one for a detection program. It depended on an individual happening to look at a query log and recognizing the anomaly.

OCR’s findings included insufficient procedures to regularly review information system activity and failure to identify and respond to suspected or known security incidents (HHS OCR). Those are the HIPAA Security Rule’s requirements for activity review and incident procedures. They are written in general terms, but their purpose is exactly what was missing: someone or something watching what privileged accounts do.

What detection would have needed to see

Large data theft leaves traces. Queries return unusual volumes of rows. Accounts act at unusual hours or from unusual hosts. Data moves to staging servers and then out of the network to destinations the organization does not normally use. Each of these signals was detectable with tools available in 2015, if the logs existed, were collected centrally and were reviewed against a baseline.

The difficulty was operational. Database query logging at scale generates large volumes. Security operations centers in 2015 were usually built around network and endpoint alerts, not database activity. Analysts saw thousands of alerts a day, and database anomalies were often not among them because no one had built the rules. Detection required deciding in advance which accounts and tables mattered most and writing rules for them.

What a detection program for privileged data access involves

The practical starting point is narrow. Identify the few data stores whose theft would be catastrophic. Turn on query and export logging for them. Send those logs to the security monitoring system. Write a small number of rules: bulk reads by human accounts, any read by an administrator account outside a change window, and transfers from database servers to new external destinations.

Then assign ownership. Someone in security operations reviews these alerts daily, with a named database contact who can confirm whether activity was legitimate. Over time, rules are tuned to remove routine noise. The measure of success is the time from an unusual bulk query to a human looking at it. In Anthem’s case, the theft ran for weeks.

Why long dwell times were common

Anthem was not unusual in 2015 in taking months to find an intruder. Security vendors’ annual incident reports from the period commonly described intrusions that lasted months before discovery. Many victims learned of breaches from outside parties such as law enforcement or banks rather than from their own monitoring. Anthem found its own intrusion, which counted in its favor. The deeper problem was industry-wide: monitoring programs were designed around malware and network signatures, while patient attackers using valid credentials looked like normal users.

What happened once it was found

The order of the response steps reported by the California regulator is instructive. Anthem shut down database access, forced a password reset across its workforce, notified the FBI and brought in an outside firm (California Department of Insurance). Public disclosure followed within a week of discovery. That was fast by the standards of the time, when some organizations took months to announce breaches.

Speed had trade-offs. Early disclosure came before the company knew the full population or the full set of data taken, so figures and details were refined over the following weeks. A company-wide password reset disrupted work and could tip off an attacker still present. Those costs were worth paying. An organization planning its own response can settle three things in advance. The first is who has authority to shut down a production database. The second is how a mass credential reset would run. The third is how early public statements will handle facts that are still uncertain. Each is far harder to decide during an incident than before one.

What a board should see

The 2020 state settlement required regular security reporting to Anthem’s board (Connecticut Attorney General). Boards in 2015 usually saw security as a list of projects and a count of incidents. The Anthem chain suggests a more useful set of measures, one for each stage. For entry, the share of workforce and subsidiary accounts protected by a second factor. For privilege, the number of accounts able to read bulk sensitive data and the share under vault control. For the store, the number of people on file who are no longer served. For detection, the time from a test bulk query to analyst review. Each is a number a board can track quarter to quarter, and each maps to a specific failure the regulators later named.

Egress controls as the last layer

Even when detection fails, outbound controls can limit what leaves. Database servers rarely need to connect to arbitrary internet destinations. Restricting their outbound traffic, and alerting on large transfers from internal staging areas, forces an attacker to work harder and creates more chances to be seen. These controls are cheap relative to their value, but they require network teams to know which servers hold sensitive data and what they legitimately talk to.

What the chain meant for different organizations

Every regulated organization in 2015 had some version of each stage: a workforce exposed to phishing, administrators with standing access, warehouses accumulating history and monitoring weak on database activity. The emphasis differed by type.

Health plans

Plans held the largest identity stores in health care, often including data about people they no longer covered and members of affiliated plans. Their priorities after Anthem were warehouse retention, removal of Social Security numbers from analytic data and privileged access management for data teams. As covered entities under HIPAA, their Security Rule obligations were direct. State insurance regulators, which examine plans’ operations, also took an interest in their security programs.

Health systems

Hospitals and physician groups held clinical records as well as identity data, and their electronic health records gave many clinicians broad access by design. Their exposure ran through clinical systems, research data stores and the population health warehouses they were building for value-based contracts. The Anthem chain applied to those warehouses directly. A system building a store of claims and clinical data for an accountable care program was creating its own version of the asset Anthem lost.

Employers and public agencies as plan sponsors

Employers that sponsored self-funded plans, including state and local governments, sent workforce data to administrators. After Anthem, their procurement and benefits teams had reason to ask administrators about retention, segregation of their data, privileged access and notification terms. Public agencies also had their own large personnel databases. Later in 2015, the breach of personnel records at the federal Office of Personnel Management showed the same pattern of credential misuse and a long dwell time in a government setting.

Business associates and vendors

Analytics vendors, claims processors and IT service providers held data for many clients. Their administrator accounts could reach multiple customers’ data. For them, the Anthem chain argued for strict tenant separation, per-client keys and access paths, and the same privileged access and logging controls their clients would expect internally. Their business associate agreements increasingly reflected these expectations.

State Medicaid agencies and public employee plans

State Medicaid agencies contract with managed care organizations, many of them owned by national insurers, and send them enrollment data on large numbers of residents. State employee health plans do the same with administrators. After Anthem, these agencies had a procurement lever. They could write data security requirements, retention limits, breach notice timelines and audit rights into managed care and administrative services contracts. Agencies that did so gained visibility into how their residents’ data was protected. Those that relied on general compliance language learned about their vendors’ controls only after an incident.

Ambulatory networks and smaller practices

Physician groups and ambulatory networks rarely held stores the size of a health plan’s. They did hold identity and insurance data for every patient, often in practice management systems run by vendors and administered by small IT teams. For them the chain compressed. A single phished account might belong to the person who also administered the practice management database. The relevant controls were the same in miniature. They were a second factor on administrative accounts, separate accounts for email and administration, vendor access limited and logged, and a check on how many years of former patients the system retained.

Banks and other financial firms watching from outside

Financial institutions were not covered by HIPAA, but they held the same identity data and faced the same chain. Their examiners already expected access controls and monitoring under the Gramm-Leach-Bliley Act safeguards guidance. Anthem was a reminder that identity data stolen from a health plan could be used to open accounts or take over existing ones at a bank. Fraud teams treated large health plan breaches as an input to their own risk models.

The case for encryption, and why it does not change the conclusion

The strongest objection to this argument holds that encryption deserved its prominence. Encrypting data at rest is a basic control. HIPAA’s breach notification rule gives a safe harbor for data encrypted to HHS’s guidance standard, which means a loss of properly encrypted data may not count as a reportable breach. An organization that encrypted everything would have one less thing to explain.

Where the objection is right

Encryption at rest is cheap relative to its value against lost devices, stolen backups and some insider threats. Laptops, removable media and backup tapes were a leading cause of reported health data breaches in the years before 2015. For those, encryption works. Field-level encryption or tokenization, where only certain services can decrypt sensitive fields, does help against a compromised administrator, because the administrator’s account does not hold the keys. In that more specific form, encryption is part of the answer.

The 2020 multistate settlement with Anthem listed encryption among the security measures it required, alongside segmentation, logging and monitoring, access controls and two-factor authentication (Connecticut Attorney General). Regulators saw no reason to choose between these controls, and neither should anyone else.

A second objection: a sophisticated attacker would have succeeded anyway

Another objection holds that a well-resourced group, of the kind often assumed to have state backing, will get in eventually, so specific controls make little difference. The Justice Department described the group as extremely sophisticated, though the indictment did not allege that it acted for a government (Justice Department). There is truth in this. A determined adversary with time and skill can defeat individual controls.

The conclusion does not follow. Every stage of the chain gave the defender a chance to raise the attacker’s cost or shorten their time inside. Privileged access controls would have forced the group to find another route. Retention limits would have shrunk what any route could reach. Database monitoring would have shortened weeks of theft to days or hours. Sophisticated attackers still choose the cheapest path, and controls change which path is cheapest. The size of the breach, unlike the fact of intrusion, was largely within the organization’s control.

Where it falls short

Disk or database encryption decrypts data for any authorized query. Applied in the usual way, it would not have stopped what happened at Anthem. Focusing public debate on it risked suggesting that an encrypted database was a safe one. Organizations that responded to Anthem by turning on database encryption, and nothing else, addressed the wrong stage of the chain.

The field-level designs that do help are access architecture decisions more than encryption decisions: which services can see which fields, under which identities and with what logging. That brings the argument back to privilege, minimization and detection.

What a security leader could check now

  • Which accounts, human and service, can read or export our largest stores of identity data, and do any use a password alone?
  • How many years of history do our warehouses hold on people we no longer serve, and who decided to keep it?
  • Would a bulk read by an administrator account outside a change window reach a human analyst within hours?
  • Can our database servers send data to arbitrary internet destinations?
  • Which of our acquired or subsidiary networks still trust the parent’s directory without meeting its baseline?

With hindsight

Anthem’s figure of 78.8 million affected people held. In March 2015 it filed its breach report with the HHS Office for Civil Rights. In October 2018 OCR announced that Anthem had agreed to pay 16 million dollars and adopt a corrective action plan, then the largest HIPAA settlement. The previous record had been 5.55 million dollars, paid in 2016. OCR found that the attackers had stolen the data of almost 79 million people between December 2, 2014 and January 27, 2015. It cited failures to conduct an enterprise-wide risk analysis, to review system activity, to identify and respond to security incidents and to implement adequate access controls (HHS OCR, October 15, 2018).

Attribution came from the Justice Department. In May 2019 a federal grand jury in Indianapolis indicted a Chinese national, Fujie Wang, and another defendant charged as John Doe. They were alleged to be members of a China-based hacking group that, beginning in February 2014, broke into Anthem and three other U.S. businesses (Justice Department, May 9, 2019).

State enforcement followed. In September 2020 a coalition of state attorneys general led by Connecticut announced a 39.5 million dollar settlement with Anthem. The settlement required an information security program based on zero trust architecture and regular security reporting to the board. It also required specific controls for segmentation, logging and monitoring, access controls, two-factor authentication, encryption and penetration testing, with third-party assessments for three years. Anthem had earlier settled consumer class litigation with a 115 million dollar fund (Connecticut Attorney General, September 30, 2020).

The settlement terms read as a list of the attack stages. Two-factor authentication and access controls addressed the privilege stage. Segmentation addressed movement from the entry point. Logging and monitoring addressed detection. The phrase “zero trust architecture” in a 2020 settlement shows how far thinking had moved from the 2015 debate about encryption toward identity and access as the center of security design.

Health care did not escape the pattern. In February 2024 a ransomware attack on Change Healthcare, a UnitedHealth Group subsidiary that processes claims for much of the industry, disrupted payments to providers across the country. UnitedHealth’s chief executive told Congress in May 2024 that the attackers had used compromised credentials on a remote access portal that lacked multifactor authentication. The company later put the number of affected individuals at about 190 million, which exceeded Anthem’s total. A single credential without a second factor had again opened a store of data on a large share of the population.

Regulation moved slowly. HHS proposed an update to the HIPAA Security Rule in January 2025 that would make multifactor authentication and encryption required rather than addressable in most cases, and require asset inventories and network maps. By mid-2026 the federal regulatory agenda had moved final action on that proposal to 2027. Organizations should check its status with counsel.

What held up from February 2015 was the attack chain. Phishing to credentials to bulk data, with a long dwell time, remained the common shape of large health data breaches. What did not hold up was the idea that encryption at rest was the decisive control. Every later enforcement action and settlement emphasized access, monitoring and multifactor authentication. Data minimization remains the least adopted of the four controls. Warehouses and data lakes have grown, and organizations still keep more history on more people than most of their analysis requires.

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

  • Anthem
  • data breach
  • privileged access
  • HIPAA
  • data minimization