145 days unpatched: how the Equifax breach actually happened
Congress documented the Equifax breach step by step: one unpatched web app, credentials on a file share, 48 databases and an expired certificate.

On 9 March 2017, Equifax emailed roughly 430 people to say that a critical vulnerability in Apache Struts was being actively exploited and had to be patched within 48 hours under company policy. The email was accurate, timely and correctly addressed. One of the systems running the vulnerable software was a consumer dispute portal built in the 1970s, and nobody patched it.
The resulting breach is probably the most thoroughly documented intrusion in corporate history, because the House Oversight Committee obtained the forensic findings and published a step-by-step account. That makes it unusually valuable: you can follow the attacker from the first command to the last file, and see exactly which five or six controls would each have stopped it.
The vulnerability
CVE-2017-5638 is a remote code execution flaw in the Jakarta Multipart parser in Apache Struts 2. In practice it let an unauthenticated attacker send a crafted HTTP header to a vulnerable web application and have the server run commands. The Apache Software Foundation published the security bulletin on 7 March 2017. Exploit code circulated within a day.
The US Computer Emergency Readiness Team notified Equifax on 8 March. Equifax's Global Threat and Vulnerability Management team passed it on internally on 9 March, held a meeting about it on 16 March, and the vulnerable installation inside the Automated Consumer Interview System was not fixed. ACIS was the internet-facing portal through which consumers dispute items on their credit file, originally built in the 1970s and still running in a data center in Alpharetta, Georgia.
The committee calculated that Equifax's systems were exploitable from 8 March until 30 July 2017: 145 days.
The intrusion, step by step
These stages come directly from the committee's report, which drew on the Mandiant forensic investigation.
- 10 March 2017: first exploitation. Mandiant found the earliest evidence of the Struts flaw being exploited, including the attacker running whoami to see what account they had landed as. This was reconnaissance rather than the main event.
- 13 May 2017: the attack proper begins. Attackers entered through the Struts vulnerability on the ACIS application servers. The environment had firewalls at the perimeter of its web servers, and exploiting the application servers bypassed them.
- Foothold. They uploaded web shells, small malicious scripts that give an attacker a command interface on a compromised web server. Around 30 unique web shells were used. Mandiant noted that file integrity monitoring could have detected their creation. ACIS did not have it enabled.
- Privilege escalation via a file share. The attackers reached a mounted file share holding unencrypted application credentials in a configuration file. Equifax's interim chief security officer told investigators that storing credentials that way was against company policy. Access was possible because Equifax did not restrict access to sensitive files across its legacy systems.
- Lateral movement into 48 databases. ACIS needed three databases to perform its function. It was not segmented from unrelated systems, so the stolen credentials worked against 48 other databases.
- Internal reconnaissance. About 9,000 queries were run. The attackers read table metadata to find out what each table held, then queried the interesting ones. 265 queries returned unencrypted personal information.
- Exfiltration. The results were written to files, compressed, placed in a web-accessible directory and pulled out using Wget, an ordinary command-line download tool. An estimated 35 IP addresses interacted with the environment.
- 29 July 2017: discovery, by accident. Equifax's countermeasures team uploaded 67 new SSL certificates to the traffic inspection appliance at the Alpharetta data center. Inspection resumed and suspicious traffic appeared immediately.
- 30 July 2017: containment. After seeing more suspicious activity, Equifax took ACIS offline. Vulnerability testing that day also found SQL injection and insecure direct object reference flaws in the ACIS code.
The expired certificate, which is the detail everyone should remember
Equifax had an intrusion detection system and an intrusion prevention system watching ACIS. Both sat behind an SSL visibility appliance whose job was to decrypt traffic so those tools could read it. The certificate on that appliance for the ACIS domain expired on 31 January 2016, according to documents produced to the committee. The device's default behavior was to let traffic pass through uninspected rather than to block it or raise an alarm.
So for 19 months, encrypted traffic to and from a credit bureau's internet-facing dispute portal flowed past two security systems that could not read it. The report notes that GAO recorded the expiry as ten months, based on different evidence, and the committee's figure came from Equifax's own master list of expired certificates. Either number makes the point.
The 76-day attack was discovered within hours of monitoring being switched back on. Equifax could not reconstruct what the attacker had done before 29 July, because the traffic had never been recorded.
Confirmed, reported and unknown
Confirmed in official reports
- The vulnerability, the notification dates, the 48-hour internal policy and the failure to patch ACIS.
- The 13 May to 30 July attack window, 76 days, and the 145-day exposure period.
- Web shells, the unencrypted credentials on a mounted file share, 48 unrelated databases, approximately 9,000 queries and 265 successful ones, exfiltration via Wget.
- The 19-month expired certificate on the traffic inspection appliance and the fail-open default.
- The absence of file integrity monitoring and of network segmentation.
- Victim counts of 143 million announced, later about 148 million; FTC's figure of approximately 147 million; GAO's at least 145.5 million.
- The July 2019 settlement of at least $575 million, up to $700 million, with the FTC, CFPB and 50 states and territories.
Reported but not established
- Attribution. Equifax employees traced suspicious traffic to IP addresses in China and to a German ISP leased to a Chinese provider. In 2020 the US Department of Justice indicted members of the Chinese military over the breach. An indictment is an allegation, and no one has been tried.
- Whether the 10 March activity and the 13 May attack were the same actor. Equifax characterized them as separate.
- What has happened to the data since. It has not appeared for sale in the way criminal breach data normally does, which is one of the arguments for the espionage theory, and remains an inference.
Unknown
- Exactly what the attackers did before 29 July 2017, because the monitoring gap destroyed the evidence.
- Why the ACIS Struts installation was missed. The company's public explanation centered on human error and communication failure, and one employee was dismissed for not forwarding an email, which the committee treated skeptically given the scale of the process failure.
Why it matters technically
The useful thing about Equifax is that it is not a story about one failure. It is six, in series, and any one of them being absent would have produced a smaller headline.
- Patch coverage depends on asset inventory. You cannot patch software you do not know you are running. The committee found Equifax did not know what software was in its legacy environments. A vulnerability alert is only as good as the list of systems you check it against.
- Segmentation caps blast radius. An application that needs three databases should be able to reach three databases. That single control would have reduced this breach by roughly 94% of the databases involved.
- Secrets in configuration files are a privilege escalation path. Plaintext credentials on a reachable share turn code execution on one server into access across an estate. Use a secrets manager, restrict the share, and alert on access to it.
- Encryption at rest matters because queries have to return something. 265 queries found unencrypted personal data. Encrypted fields with keys the application does not casually hold would have changed what those queries produced.
- Monitoring fails silently. This is the most under-appreciated lesson. Nobody noticed that a security control had been blind since January 2016 because blind controls generate no alerts. Monitor your monitoring: certificate expiry, agent health, log flow volumes, and the absence of events where you would expect some.
- Fail-open versus fail-closed is a decision someone makes. The appliance passed traffic when it could not inspect it. That is sometimes the right trade-off, and it needs to be deliberate and alerted on.
Our articles on vulnerability management and zero days, least privilege and how data breaches unfold cover these in a more general form.
What an ordinary business should take from it
- Know what you run that faces the internet, including the old things. Write the list down, including framework and library versions, and check it against advisories.
- Close the loop on patch alerts. An email to 430 people is a notification, not a process. Someone must confirm each affected system is fixed, and the exceptions must be tracked.
- Separate your internet-facing applications from the rest of your network. If your website's server can reach your finance database, fix that before buying anything new.
- Hunt for secrets in your own files. Credentials in config files, scripts and spreadsheets are the most common privilege escalation route in small organizations too.
- Test that your security tools are seeing traffic. Pick one: when did you last confirm your logging actually receives events from every system you think it covers?
Got a message about a breach settlement or credit monitoring? Equifax-themed scams ran for years after the breach. Paste the message into CyberWatch AI Scan for a free check.
If your data was in it
With roughly 147 million people affected, most US adults should assume their Social Security number has been exposed at some point, by this breach or another. The durable protections are a credit freeze at all three bureaus, strong unique passwords with multi-factor authentication on financial accounts, and skepticism toward anyone who calls knowing your details. Our guides to checking whether you are in a breach, reporting identity theft and account security cover the practical steps.
The short version
A critical vulnerability was announced, flagged internally, and left unpatched on an internet-facing portal for 145 days. Attackers walked in, found plaintext credentials on a file share, reached 48 databases the application had no business touching, ran 9,000 queries, and copied out the personal data of around 148 million people over 76 days while the security tools watching that network segment sat blind behind a certificate that had expired a year and a half earlier.
The committee called it entirely preventable, and the published detail makes that hard to dispute. Every control that failed here is one an ordinary organization can check this week, which is the reason this incident still earns its place in a security briefing nearly a decade on. For a structured starting point, see our business cybersecurity guide.
Frequently asked questions
What caused the Equifax breach?
Attackers exploited CVE-2017-5638, a critical vulnerability in the Apache Struts web framework, in Equifax's internet-facing consumer dispute portal known as ACIS. The patch had been available since 7 March 2017 and Equifax's own policy required critical patches within 48 hours, but the ACIS installation was never patched.
How long were attackers inside?
The House Oversight Committee's report states the attack ran from 13 May to 30 July 2017, a period of 76 days. Separately, the first evidence of exploitation of the Struts flaw at Equifax dates to 10 March 2017, and the systems remained vulnerable for 145 days.
How did the attackers get from one web application to 48 databases?
After installing web shells, they found a mounted file share containing unencrypted application credentials in a configuration file. ACIS needed access to three databases to do its job but was not segmented from the rest of the estate, so those credentials opened 48 unrelated databases.
Why did nobody notice the data leaving?
The device that decrypted and inspected traffic to and from ACIS had an expired SSL certificate. Per documents obtained by the committee, it expired on 31 January 2016, leaving 19 months with no visibility. The device's default setting let traffic through uninspected rather than blocking it.
How many people were affected?
Equifax announced 143 million consumers on 7 September 2017. The figure later grew to about 148 million. The FTC's settlement announcement describes approximately 147 million people, and GAO reported at least 145.5 million. The differences reflect different counts at different dates.
What did the attackers actually do inside the databases?
They ran approximately 9,000 queries. 265 of those successfully located unencrypted personal information, which they wrote to files, compressed, moved to a web-accessible directory and pulled out using the Wget utility. About 30 web shells and an estimated 35 IP addresses were involved.
What were the consequences for Equifax?
The chief information officer and chief security officer retired days after the announcement and the chief executive left later that month. In July 2019 Equifax agreed to pay at least $575 million, and potentially up to $700 million, in a global settlement with the FTC, the CFPB and 50 states and territories.
Was it preventable?
The House committee's conclusion was that the breach was entirely preventable. Patching, file integrity monitoring, network segmentation, credential hygiene and certificate management were each sufficient on their own to have changed the outcome.
What is the most transferable lesson?
Patching is necessary and not sufficient. The reason one unpatched application became 148 million records was everything behind it: flat network access, plaintext credentials on a file share, unencrypted data at rest and monitoring that had been silently off for over a year.
Sources
- The Equifax Data Breach: Majority Staff Report, US House Committee on Oversight and Government Reform
- Data Protection: Actions Taken by Equifax and Federal Agencies in Response to the 2017 Breach (GAO-18-559), US Government Accountability Office
- Equifax to Pay $575 Million as Part of Settlement with FTC, CFPB, and States Related to 2017 Data Breach, US Federal Trade Commission
- CVE-2017-5638 Detail, NIST National Vulnerability Database
- CVE Record: CVE-2017-5638, CVE Program
- Apache Struts 2 Security Bulletin S2-045, The Apache Software Foundation
- Exploit Public-Facing Application, Technique T1190, MITRE ATT&CK
- Server Software Component: Web Shell, Technique T1505.003, MITRE ATT&CK
- Chinese Military Personnel Charged with Computer Fraud, Economic Espionage and Wire Fraud for Hacking into Credit Reporting Agency Equifax, US Department of Justice
- Equifax data breach FAQ: What happened, who was affected, what was the impact, CSO Online


