Equifax: The Unpatched Server That Exposed 148 Million People
The patch was released on March 7, 2017. Equifax was alerted the next day and told its own teams to patch within 48 hours. One internet-facing system — a dispute portal first built in the 1970s — never got the patch, and a scan a week later missed it. Attackers used that opening for 76 days, ran about 9,000 queries across 48 databases, and left with the personal information of 148 million people, while the device meant to watch that traffic sat blind behind an expired certificate. The House Oversight Committee called the breach 'entirely preventable.' This is the documented story, from the GAO report and the House investigation, of how it happened and what changed afterwards.
The short answer
On March 7, 2017, a critical flaw in Apache Struts — a widely used framework for building web applications — was publicly disclosed, and a patch (a software fix) was released the same day. Equifax, one of the largest consumer reporting agencies in the world — companies that collect people's credit information and sell reports about it — was alerted the next day, and on March 9 its security team emailed more than 400 people instructing anyone running Struts to patch within 48 hours. One system never got patched: the Automated Consumer Interview System (ACIS), Equifax's internet-facing portal where consumers dispute errors in their credit reports, a custom-built system first developed in the 1970s. A company scan on March 15 did not detect the flaw on any external system. On May 13, attackers entered through ACIS, planted web shells (small backdoor programs that give remote control through a web browser), found a file containing unencrypted usernames and passwords, and used those credentials to reach 48 unrelated databases. Over 76 days they ran about 9,000 queries and removed personal data in small, encrypted increments that blended in with normal traffic — while the device meant to inspect that traffic was inactive because its security certificate had expired. Equifax discovered the attack on July 29 only after that certificate was renewed. The House Oversight Committee's December 2018 report put the final total at 148 million people and called the breach "entirely preventable."
The company: what Equifax did
Equifax is a consumer reporting agency (CRA) — also called a credit bureau. CRAs gather information about consumers from the companies they do business with (banks, lenders, card issuers and others, called furnishers), analyse it to create credit reports and scores, and sell those reports to third parties. Lenders use them to decide whether to grant a loan and at what rate; insurers, employers and utilities use them too. Federal agencies — the Internal Revenue Service (IRS), the Social Security Administration (SSA) and the U.S. Postal Service (USPS) — were also Equifax customers, buying identity-verification services.
Two features of that business model matter for this story. First, as the House report stresses, consumers do not voluntarily provide their information to CRAs and cannot opt out of the collection. People never chose to hand Equifax their names, Social Security numbers, birth dates and addresses — the data arrived whether they liked it or not. Second, that makes a CRA, in the report's words, a high-value target with a heightened responsibility to protect what it holds.
Equifax was one of the three national CRAs in the United States, alongside Experian and TransUnion. From 2005, under Chief Executive Officer Richard Smith, it pursued what the House report calls an aggressive growth strategy — acquiring companies, IT systems and data. The growth worked for the bottom line, and it also built a complex, ageing IT estate. Three weeks before the breach was announced, Smith boasted that Equifax was managing "almost 1,200 times" the amount of data held in the Library of Congress every day. The House report's verdict on the other side of that growth: Equifax "failed to implement an adequate security program to protect this sensitive data."
The problem: four failures that lined up
Equifax's own investigation, summarised by the U.S. Government Accountability Office (GAO) in report GAO-18-559 (August 2018), identified four major factors that let the attackers succeed: identification, detection, segmentation, and data governance. Each one is simple on its own. Together, they removed every layer that should have stopped this attack.
Identification — the patch never reached the system. After the March 2017 alert, Equifax circulated the vulnerability notice among its system administrators. GAO reports what Equifax officials told it: the recipient list for the notice was out of date, so the notice was not received by the individuals who would have been responsible for installing the patch on the dispute portal. Then the safety net failed too: Equifax scanned its network a week after the vulnerability was identified, and the scan did not detect the vulnerability on the online dispute portal. A patch process that depends on an email list, plus a scanner that cannot see the system it most needs to see, is not a patch process — it is a hope.
What this shows: the fix existed, the warning was sent, and the deadline was set — the chain broke at one system nobody patched and one scan that could not see it. Everything after that was consequences.
Detection — the watchdog was asleep, and nobody checked. Equifax had installed a device to inspect network traffic for evidence of malicious activity. A misconfiguration caused by an expired digital certificate — a certificate is an electronic token a system uses to prove identity and to inspect encrypted traffic — meant encrypted traffic passed through without being inspected. On the duration, the two official reports differ, and this page keeps them separate: GAO (August 2018) said the certificate had expired about 10 months before the breach occurred; the House report (December 2018), reviewing the underlying documents, said the device used to monitor ACIS network traffic had been inactive for 19 months due to the expired certificate, and noted the discrepancy itself. The House report also found that Equifax had allowed over 300 security certificates to expire in total, including 79 certificates for monitoring business-critical domains. An expired certificate does not raise an alarm by itself — that is exactly why expiry has to be tracked as its own job.
Segmentation — one door opened the whole building. Segmentation means dividing a network into isolated sections, so that breaking into one section does not grant access to the rest. GAO's summary is blunt: because individual databases were not isolated or segmented from each other, the attackers were able to access additional databases beyond the ones related to the online dispute portal. The portal was a public-facing dispute tool. Behind it, reachable from it, sat database after database of credit data.
Data governance — the keys were in a file, unencrypted. Data governance includes setting limits on who — and what — can access sensitive information, including credentials. The attackers found unencrypted usernames and passwords that gave them access to additional databases, and used them to expand from the 3 databases associated with the dispute portal to 48 more. And there was a fifth gap GAO lists alongside the four: no restriction on how often databases could be queried, which let the attackers execute approximately 9,000 queries — many more, GAO notes, than normal operations would need. Nobody was counting.
An analogy: a warehouse learns that one model of door lock can be picked, and a free replacement lock is available. The notice about the lock is sent to an old mailing list, so the crew responsible for one side door never hears. An inspector walks the building and misses that door. The side door also turns out to open into every storage room, a spare set of master keys sits unlabelled in a drawer just inside it, and the security camera covering that corridor has had a dead battery for months. Any one fix breaks the chain. None was made.
What happened in 2017
All dates are 2017 unless stated. The timeline is the House report's "Timeline of Key Events," cross-checked against GAO-18-559.
March 7 — The Apache Struts Project Management Committee announces vulnerability CVE-2017-5638 and releases the patch. The flaw allows an attacker to execute commands on affected systems — in plain English, to make the server run instructions of the attacker's choosing.
March 8–9 — The United States Computer Emergency Readiness Team (US-CERT) sends Equifax an alert. On March 9, Equifax's Global Threat and Vulnerability Management team emails the alert to over 400 people, instructing anyone running Apache Struts to apply the patch within 48 hours, and the team holds a meeting about the vulnerability on March 16. ACIS is not patched.
March 10 — The first probing arrives almost immediately. GAO reports that unidentified individuals scanned Equifax's systems to determine whether they were susceptible to the vulnerability US-CERT had identified two days earlier, discovered the dispute-portal server running the vulnerable software, gained access, and confirmed they could run commands. No data was taken at this time.
March 15 — Equifax's security team runs scans to identify any systems containing the Struts vulnerability. The scans do not detect the vulnerability on any externally facing system. The exposed portal — the one attackers had already found five days earlier — remains exposed.
May 13 — In what GAO describes as a separate incident following the initial access, the main attack begins. Attackers enter through the Struts vulnerability in ACIS and drop web shells to obtain remote control. Mandiant, the forensic firm Equifax later hired, concluded the attackers compromised two systems supporting the ACIS portal and used thirty unique web shells during the intrusion. Inside, they find the file of unencrypted credentials and pivot — move sideways — out of the ACIS environment into 48 unrelated databases.
May 13 – July 30 — For 76 days, the attackers work quietly. They send about 9,000 queries to those 48 databases, successfully locating unencrypted personally identifiable information (PII) — data that identifies a person — 265 times, per the House report. GAO describes how they hid: they used existing encrypted communication channels connected to the portal to send queries and retrieve data, so their activity blended in with regular network activity, and they removed the stolen data in small increments using standard encrypted web protocols, disguised as normal traffic. The one device that should have inspected that encrypted traffic was inactive behind its expired certificate.
What this shows: for 76 days the attackers walked from database to database, running about 9,000 queries across 48 databases. The device meant to inspect that traffic was blind the whole time — its certificate had expired, so nobody saw a thing until the certificate was renewed.
July 29 — The attack is discovered almost by accident, in the course of fixing the blind spot. GAO says a network administrator conducting routine checks of the operating status and configuration of IT systems found the misconfigured equipment; Equifax renewed the expired certificate, inspection of network traffic resumed, and the administrator immediately recognised signs of an intrusion — system commands being executed in ways that were not part of normal operations. The House report adds that the suspicious traffic was identified as coming from an IP address originating in China, and that Equifax blocked the suspicious traffic.
July 30 — Equifax observes additional suspicious activity, including traffic from a second IP address owned by a German internet provider but leased to a Chinese provider. Equifax takes the ACIS portal offline for emergency maintenance. The attack ends when ACIS goes offline — 76 days after it began.
July 31 – August 2 — Equifax determines PII may have been exfiltrated, assigns the incident-response effort the code name Project Sierra, and informs Chief Executive Officer Richard Smith on July 31. On August 2 it engages the cybersecurity firm Mandiant to conduct a forensic investigation, contacts outside counsel, and notifies the Federal Bureau of Investigation.
September 4–7 — Equifax and Mandiant complete a list of 143 million affected consumers. On September 7, Equifax publicly announces the breach, stating that the information accessed included names, Social Security numbers, birth dates, addresses, driver's licence numbers, credit card numbers, and dispute documents.
The damage
The headline number kept growing as the forensic work continued, which is itself part of the damage story: Equifax could not quickly say who was affected, because it first had to reconstruct what the attackers had done. The September 7 announcement said 143 million U.S. consumers. On October 2, Mandiant completed its forensic investigation and Equifax revised the figure up by 2.5 million, to 145.5 million — GAO's August 2018 report uses "at least 145.5 million" in the United States, plus nearly 1 million consumers outside it. On March 1, 2018, Equifax identified a further 2.4 million people whose names and partial driver's licence information were taken. The House report's final figure, December 2018, is 148 million — nearly half the U.S. population and 56 percent of American adults. Where the reports differ, this page uses the later House figure as the final count and dates GAO's figure to August 2018.
What was taken matters as much as how much. Per Equifax's September 7 announcement as recorded in both reports: names, Social Security numbers, birth dates, addresses and driver's licence numbers. GAO adds the more specific counts Equifax determined: credit card numbers for approximately 209,000 consumers, and certain dispute documents containing PII for approximately 182,000 consumers — documents people had submitted as evidence in credit disputes, such as utility bills. The House report's sixth recommendation states the sharpest consequence plainly: attackers stole the Social Security numbers of an estimated 145 million consumers, so nearly half the country's Social Security numbers are no longer confidential. A password can be changed. A Social Security number, for practical purposes, cannot.
The cost to Equifax itself, from figures the House report takes from the company's own third-quarter 2018 results: $221.5 million in costs related to the cybersecurity incident through the first nine months of 2018 — $193.2 million of technology and data security spending, $61.4 million of legal and investigative fees, and $11.9 million of product liability, partly offset by $45.0 million of insurance recoveries. The company's interim chief executive said in November 2017 that Equifax had increased security spending fourfold since the breach was discovered; for scale, the security team's annual budget at the time its chief left in September 2017 was $38 million, per testimony recorded by the House report.
The leadership cost was total at the top. The Chief Information Officer and the Chief Security Officer both took early retirements on September 15, eight days after the announcement. Chief Executive Richard Smith left on September 26. On October 2, Equifax terminated a senior vice president who managed a number of IT systems including ACIS, for failing to forward the email regarding the Struts vulnerability — a detail that sits uneasily beside GAO's finding that the distribution list itself was out of date, and both accounts are part of the official record.
And the damage landed, finally, on the people the data described. When Equifax announced the breach, it was, in the House report's words, unprepared to support affected consumers: the dedicated breach website and call centres were immediately overwhelmed, and consumers could not obtain timely information about whether they were affected or how to protect themselves — despite Equifax standing up a call centre staffed by 1,500 temporary employees as part of its notification effort, code-named Project Sparta. The federal customers reacted as customers do: IRS, SSA and USPS all conducted their own assessments and visited Equifax's Alpharetta, Georgia data centre; the IRS terminated one of its contracts with Equifax; and the Bureau of Consumer Financial Protection and the Federal Trade Commission opened an investigation in September 2017.
How they recovered
Step one: stop the bleeding. On July 29, once the renewed certificate made the traffic visible, Equifax blocked the suspicious traffic. When suspicious activity continued on July 30, it took the ACIS portal offline entirely — the step that ended the attack. Containment here was not subtle: the vulnerable door was closed by removing the door.
Step two: investigate under a code name. The incident-response effort ran as Project Sierra from July 31. Mandiant's forensic investigation ran from August 2 to October 2, 2017. Because the attackers' commands had been recorded in electronic logs that were not damaged or erased, investigators could reconstruct the sequence of actions — which queries ran, and which data they returned.
Step three: recreate the attack to count the victims. GAO describes the method: Equifax recreated the attackers' database queries on a separate system that could run them at high speed, reconstructed which queries had successfully extracted data, and matched the incomplete records in the stolen data against its internal databases to identify the individuals involved — removing duplicates and linking partial records by Social Security number and name. That work produced the 143 million figure announced on September 7, and its continuation produced both later revisions.
Step four: notify — badly at first. Under Project Sparta, Equifax built a website for individuals to check whether they were affected and to register for credit monitoring and identity theft services, and staffed call centres with 1,500 temporary employees. Both were immediately overwhelmed on announcement day. Equifax offered free credit monitoring and protection services for one year to any consumer who requested it, per the House report. The Department of Homeland Security offered assistance with the response; GAO reports that Equifax declined it because it had already retained an external cybersecurity consultant.
The fix: technology and controls that came out of it
The clearest fix list in the record is Mandiant's. Its September 19, 2017 report contained eleven remedial recommendations, and in August 2018 Mandiant and Equifax officials confirmed to congressional staff that Equifax had implemented all eleven:
Patching and scanning. (1) Enhance vulnerability scanning and patch management processes and procedures. GAO's account of the system-level measures matches: a new management process to identify and patch software vulnerabilities and confirm that vulnerabilities had been addressed — the confirmation step whose absence in March 2017 is the whole story — plus a new endpoint security tool to detect misconfigurations and automatically notify administrators of identified vulnerabilities, and the upgrading or elimination of vulnerable legacy systems.
Segmentation and access. (4) Enhance network segmentation to restrict access from internet-facing systems to backend databases and data stores; (3) increase restrictions and controls for accessing data in critical databases; (8) accelerate deployment of a privileged account management solution — a system for controlling powerful administrative accounts, the grown-up version of "do not leave unencrypted passwords in a file"; and (2) reduce the scope of sensitive data retained in backend databases, so there is less to steal in the first place. GAO likewise reports additional controls at the external boundary and further restrictions on traffic between internal servers, under a new security controls framework with tighter controls on accessing specific systems, applications and networks.
Visibility. (5) Deploy additional web application firewalls — filters that inspect web traffic for attack patterns — with tuned signatures; (6) accelerate file integrity monitoring on application and web servers (watching for unauthorised changes to files, such as thirty web shells appearing); (7) enforce additional network, application, database and system-level logging; (9) enhance visibility for encrypted traffic by deploying additional inline decryption capabilities — the direct fix for the blind monitoring device; and (10) deploy additional endpoint detection and response agents, plus (11) additional email protection and monitoring. GAO reports the organisational version of the same fix: new policies and new tools to ensure network traffic is monitored continuously.
Governance, made structural. Equifax created a Chief Information Security Officer role reporting directly to the Chief Executive Officer — before the breach, the Chief Security Officer had reported to the Chief Legal Officer — with a governance structure to communicate risk regularly to the board and senior management. On June 25, 2018, Equifax also agreed to a Consent Order with regulatory agencies from eight states, committing to a board-approved written risk assessment within 90 days, a formal internal audit programme within 30 days, and specific patch-management reforms: a comprehensive IT asset inventory, a formal process to routinely identify necessary patches, an action plan for decommissioning legacy systems, and a formalised Patch Management Policy. Note what that list starts with: an inventory. You cannot patch, scan or segment systems you have not counted.
The wider lessons Congress drew. The House report's seven recommendations reach beyond one company: more transparency for consumers about what CRAs hold; a review of whether the FTC has sufficient oversight and enforcement authority; a review of how long identity monitoring should last after a breach; more transparency about cyber risk in the private sector; clear cybersecurity requirements for federal contractors; reducing the use of Social Security numbers as personal identifiers; and — aimed at every company storing sensitive data — transitioning away from legacy IT to modernised IT security solutions, because, in the report's words, the complexity of the legacy environment hosting ACIS made securing it harder.
What this means for you
Enterprise network engineers
Strip away the credit-bureau specifics and this is an enterprise story: a legacy internet-facing application, a patch notice that never reached its owners, a scanner with a blind spot, an expired certificate nobody owned, and a flat internal network behind it. Your equivalents are already in your estate. The portable checks are unglamorous and exact: maintain an asset inventory you would bet the company on, because the March 15 scan failed on inventory, not effort; verify patches landed, on the named system, after the deadline — Equifax's fix, per GAO, added exactly that confirmation step; treat certificate expiry as an outage of your security tooling, with expiry dates inventoried and alarmed like any other failure; and segment so that your most public system cannot reach your most sensitive data — the dispute portal should never have been able to query 48 unrelated databases, and your web tier should not be able to either.
Hyperscaler engineers
At fleet scale, Equifax's failures become automation questions. An out-of-date email list is, at your scale, an ownership database that has drifted from reality — and every alert routed by it fails silently in the same way. A scanner that misses one legacy system is a coverage metric that reports the fleet it knows about, not the fleet that exists. The query-count lesson scales too: 9,000 queries over 76 days is a slow leak, invisible to thresholds set for spikes; rate and volume baselines per service, with alerts on sustained abnormal totals, are how that pattern gets caught. And the certificate lesson is the sharpest: a monitoring system that fails silent is worse than none, because everyone believes they are watched. Monitor the monitors — expiry, ingestion gaps and decryption failures should page like the incidents they hide.
Try it yourself
Two of Equifax's four factors are things you can practise spotting with Packet Path's free tools. Segmentation lives or dies in firewall and access-control rules: the free Firewall / ACL Analyzer checks rule sets for shadowed, redundant and overly broad rules — the rule-level version of asking "should this system really be able to reach that one?" And identification failures are configuration drift by another name: the free Network Config Verifier compares device configurations side by side and flags mismatches, building the habit Equifax's process lacked — do not assume the fleet matches the paperwork. Compare it, every time.
Sources
Primary source — U.S. Government Accountability Office, Data Protection: Actions Taken by Equifax and Federal Agencies in Response to the 2017 Breach, GAO-18-559 (August 2018): the four factors (identification, detection, segmentation, data governance), the out-of-date recipient list, the March 10 scanning and May 13 attack as Equifax reported them, the approximately 9,000 queries and the move from 3 portal databases to 48 additional databases, the expired certificate described as expired about 10 months before the breach, the July 29 discovery during routine checks, the query-recreation method used to identify affected individuals, the 209,000 credit card numbers and 182,000 dispute documents, the revision to 145.5 million on October 2, 2017, and the remediation measures Equifax reported. https://www.gao.gov/products/gao-18-559
Primary source — U.S. House of Representatives, Committee on Oversight and Government Reform, Majority Staff Report, The Equifax Data Breach (December 10, 2018), from the Committee's 14-month investigation reviewing over 122,000 pages of documents: the "entirely preventable" finding, the day-by-day timeline (March 7 disclosure, March 8 US-CERT alert, March 9 email to over 400 people with a 48-hour instruction, March 15 scan, May 13 attack start, 76 days, July 29 certificate renewal and July 30 ACIS shutdown, September 7 announcement), the 48 databases and 265 successful PII locations, the monitoring device inactive for 19 months and over 300 expired certificates including 79 for business-critical domains, the final figure of 148 million, Project Sierra and Project Sparta, the leadership departures, Mandiant's eleven remedial recommendations and their confirmed implementation, the June 25, 2018 eight-state Consent Order, the $221.5 million cost figure through the first nine months of 2018, and the Committee's seven recommendations. https://oversight.house.gov/report/committee-releases-report-revealing-new-information-on-equifax-data-breach
- The patch existed for two months before the main attack: the failure was identification — an out-of-date email list, per GAO, meant the notice never reached the people responsible for the dispute portal, and a March 15 scan missed the vulnerable system.
- One unpatched portal was enough because the network was not segmented: attackers found unencrypted credentials and reached 48 unrelated databases, running about 9,000 queries and finding unencrypted personal data 265 times.
- Detection failed silently: the device monitoring the portal's traffic was inactive because of an expired certificate — GAO said the certificate had expired about 10 months before the breach; the House report, reviewing the documents, put the monitoring device's inactivity at 19 months.
- The final count grew over time — 143 million announced on September 7, 2017, 145.5 million by October 2, and 148 million in the House report's final figure — nearly half the U.S. population.