What happened
Replaying the Capital One March 2019 data breach
- 1
Scanning, allegedly behind VPN and TOR
Up to Mar 12, 2019A scanner looks for misconfigured firewalls on public-facing servers; the first attempt reaches Capital One through a VPN address.
- 2
Through the firewall to the role credentials
On or before Mar 22, 2019The misconfigured firewall relays outside commands to the server; the first command obtains the credentials of the server's IAM role.
- 3
Stolen credentials accepted from the internet
Mar 22 and Apr 21, 2019The role's credentials sign Amazon S3 requests from VPN addresses and from addresses believed to be TOR exit nodes; the List Buckets Command returns more than 700 names.
- 4
Copied to a home server
Mar 22, 2019The Sync Command copies the buckets the role can read, credit card application files among them, to a server at the accused's home; the access also enables decrypting data.
Credit card application and customer data of approximately 100 million people in the United States and approximately 6 million in Canada copied, about 140,000 Social Security numbers and about 80,000 bank account numbers among it; an $80 million OCC penalty and a $190 million customer settlement followed.
Where cloud-provider controls fit
IMDSv2, set on the instance or required by an organization policy, answers only a session that starts with a PUT request, which narrows which relays reach the role credentials; the sources do not confirm that the credentials came from the metadata service or say what the relay could send. A service control policy with network conditions binds only the member accounts of an organization, and the sources do not say which kind of account held the role. A resource control policy on the buckets and an S3 endpoint policy would have kept the reach of those credentials inside your VPC; a permissions boundary would have capped the role at its own buckets, and the same perimeter on AWS KMS would have delivered ciphertext. A web ACL with managed rule groups would have narrowed who could probe the server and refused the request patterns it lists, though the scanner's addresses and the request pattern are not in the sources. The firewall rule set itself stays your configuration on every cloud.
Could this have been prevented?
In hindsight, no control would have stopped it. 13 controls would have limited how far it went. The counts below are control assessments, not attack stages: several can address the same step. The same replay with the controls in place before the incident shows where each phase would have been refused or bounded.
- 0
- Prevented
- 13
- Contained
- 0
- Not covered
How it could have been prevented
The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.
Replaying the Capital One March 2019 data breach
1 · Scanning, allegedly behind VPN and TOR
2 Contained6 on other clouds2 · Through the firewall to the role credentials
4 Contained6 on other clouds3 · Stolen credentials accepted from the internet
3 Contained6 on other clouds4 · Copied to a home server
4 Contained6 on other cloudsKey facts
- An IPredator VPN address attempted to reach the data on March 12, 2019; the List Buckets and Sync commands ran on March 22 from VPN addresses and from addresses believed to be TOR exit nodes
- The List Buckets Command returned more than 700 bucket names, a command the role did not invoke in the ordinary course of business
- Approximately 100 million people in the United States and approximately 6 million in Canada were affected; about 140,000 Social Security numbers and about 80,000 linked bank account numbers were among the data
- An external researcher reported the leaked bucket list on July 17, 2019; Capital One fixed the configuration on July 19 and the FBI seized devices holding a copy of the data on July 29
- AWS attributed the intrusion to a misconfigured web application firewall; AWS believes server-side request forgery was used, and the path to the metadata service is the researchers' inference
- The OCC assessed an $80 million penalty in August 2020 for risk assessment failures before the cloud migration; the DOJ states that customer lawsuits settled for $190 million
Sources
Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.
- 1.Capital One Announces Data Security IncidentCapital One · Jul 29, 2019 · Vendor reportDiscovery, access dates, data types, counts, encryption and tokenization, cost estimate, arrest
- 2.Complaint for Violation of 18 U.S.C. § 1030(a)(2), United States v. Paige A. Thompson, No. MJ19-0344 (W.D. Wash.)U.S. Department of Justice, U.S. Attorney's Office for the Western District of Washington · Jul 29, 2019 · AdvisoryTechnical chain, logged dates, TOR and VPN use, bucket count, disclosure email, counts at the time, evidence trail
- 3.Seattle Tech Worker Arrested for Data Theft Involving Large Financial Services CompanyU.S. Department of Justice, U.S. Attorney's Office for the Western District of Washington · Jul 29, 2019 · NewsArrest, charge, cause wording, discovery dates, seizure
- 4.Former Seattle Tech Worker Indicted on Federal Charges for Wire Fraud and Computer Data TheftU.S. Department of Justice, U.S. Attorney's Office for the Western District of Washington · Aug 28, 2019 · NewsIndictment, other victims, scanner, cryptojacking, no evidence of dissemination
- 5.Superseding Indictment, United States v. Paige A. Thompson, No. CR19-0159 (W.D. Wash.), Document 102U.S. Department of Justice, U.S. Attorney's Office for the Western District of Washington · Jun 17, 2021 · AdvisoryScheme, manner and means, charged dates, other victims, counts
- 6.Former Seattle tech worker convicted of wire fraud and computer intrusionsU.S. Department of Justice, U.S. Attorney's Office for the Western District of Washington · Jun 17, 2022 · NewsVerdict, counts, trial facts, settlement figure
- 7.OCC Assesses $80 Million Civil Money Penalty Against Capital OneOffice of the Comptroller of the Currency · Aug 6, 2020 · NewsPenalty, regulatory basis, cloud migration risk assessment finding
- 8.Consent Order, In the Matter of Capital One, N.A. and Capital One Bank (USA), N.A., AA-EC-20-49 (#2020-037)Office of the Comptroller of the Currency · Aug 5, 2020 · AdvisoryControl deficiencies, 2015 cloud migration, internal audit, remediation requirements
- 9.What We Can Learn from the Capital One HackKrebs on Security · Aug 2, 2019 · ResearchSSRF account, WAF product, metadata service, AWS statement, expert commentary
- 10.A Technical Analysis of the Capital One HackCloudSploit · Aug 2, 2019 · ResearchMetadata service mechanics, credential use from outside AWS, S3 commands, mitigations
- 11.A Case Study of the Capital One Data Breach (Revised), Working Paper CISL# 2020-16MIT Sloan, Cybersecurity Interdisciplinary Systems Laboratory · Mar 1, 2020 · ResearchAttack step reconstruction, ATT&CK mapping, cloud migration context, data volume claim
- 12.Add defense in depth against open firewalls, reverse proxies, and SSRF vulnerabilities with enhancements to the EC2 Instance Metadata ServiceAmazon Web Services, AWS Security Blog · Nov 19, 2019 · AdvisoryIMDSv2 mechanics, enforcement controls, threats addressed
- 13.Amazon Web Services finds no 'significant issues' at other companies allegedly breached by Paige ThompsonCyberScoop · Aug 16, 2019 · NewsAWS account of the cause, SSRF acknowledgement, other customers, AWS commitments
The estate in the replay is illustrative and drawn at the level of attack stages; the victim's real environment is not public. Verdicts describe what a control does when it is in place before the incident. They are not an assessment of the victim's security program.