On 4 February 2025 DigiCert moved an internal business system out of a VPN-only network segment and into one reachable from the internet. Nobody reviewed the change. The system archived what customers hand a certificate authority to prove who they are: business contact details and, for a small minority, a passport, driving licence or utility bill. It also carried an account manager built for other products, with a self-registration page that did not ask for MFA.

Sixteen months later strangers had made accounts on it. DigiCert saw them on 18 June 2026, decided they could not see anything, and closed the door to new sign-ups. From 27 June to 2 July one of those accounts read archived customer data. DigiCert worked that out on 19 October. The certificate side of the business was never touched; the interesting failure is in the triage.

[INFO]
Primary source: DigiCert's own incident report in Mozilla's CA compliance tracker (Bugzilla 2084460, preliminary 20 October 2026, full report and comments to 6 November). Root programs require a CA to report incidents there in public, which is why this breach has a day-by-day timeline when most do not. Press coverage from BleepingComputer and SecurityWeek (5 November). DigiCert has not said how many people are affected.
◈ interactive artifact
Account Issue or Data Event?
The business portal's admin screen through DigiCert's own timeline: staff accounts, then strangers with no application and no MFA. See what the team believed beside what was true, and on 18 June open one of the accounts instead of only blocking sign-ups.

//What Was Exposed, and What Was Not

Per the report, most affected records hold business contact information: names, business email addresses, phone numbers and business addresses. "A small minority" also held a customer-supplied identity document showing a name and address. These are the documents a CA collects for organisation and extended validation, where it has to establish that a company and the person ordering the certificate are real. The account was read-only.

DigiCert states that "No certificate operations were impacted": nothing was issued, modified or revoked, and the system is separate from the CA, CertCentral and DigiCert ONE. BleepingComputer quotes it as saying the account "could not modify records, approve certificate validations, or issue certificates". No certificates needed replacing. Nothing in the report suggests otherwise, and the root programs are reading the same document.

//Three Contributing Factors

DigiCert lists them itself, and each is a common way for an internal system to end up on the internet.

A network change with no security review. The February 2025 migration moved the system to an "externally accessible availability zone". The application had been written for a network where only staff could reach it, so its self-registration page was harmless until the network around it changed.
Monitoring that stayed where the system used to be. Alerting was not reassessed after the move, so new accounts and request volume set off nothing.
An authorisation model with a hole in it. Access was supposed to follow from being assigned to an application. A self-registered account had no application, and reviewers concluded from that it had no access. It did: per-endpoint checks were not enforcing the assignment. The fix, released on 2 July, makes those endpoints return 401 to any account with no assigned application.

//The Nine Days

This is our reading of DigiCert's timeline, not DigiCert's.

On 18 June the team saw non-DigiCert accounts, concluded they had no data access, and put the system behind a WAF that blocked external account creation. That stops new accounts. It does nothing about accounts that already exist, and the access window DigiCert later established opens on 27 June, nine days after that block. Then, in order: 21 June, the managed security research program reports that outsiders can create accounts; 24 June, DigiCert reproduces it; 30 June, root cause found; 2 July, the authorisation patch closes the window. The data access happened while DigiCert was actively working the issue, because the issue it was working was "who can sign up", not "what can the accounts that signed up do".

DigiCert's own summary, as BleepingComputer quotes it: "The incident was initially handled as an account-creation issue rather than a data-access event." The question that would have changed the outcome on 18 June was cheap: log in as one of the strange accounts, or read its request history, and see what it can reach. The report also leaves open why the 27 June access went unnoticed until 12 October; alerting on file access and request volume was only added on 15 October. A participant in the Bugzilla thread asked what triggered the October suspicion; no answer is posted.

The system was reachable from February 2025. The report confirms access only in the June to July window. It does not say whether logs from the earlier sixteen months exist or were reviewed, so that period is neither cleared nor implicated.

//Why the Data Is Useful to Someone

A list of the people at each company who buy certificates, with their direct email and phone, is a ready-made pretext. A message that says your organisation validation is about to lapse and asks for a document or a login is exactly what these people expect to receive from a CA. Anyone who handles certificate ordering at a DigiCert customer should treat unexpected revalidation or renewal requests with suspicion and go to CertCentral directly rather than through a link. The identity documents are the more serious exposure for the individuals concerned; DigiCert says it is notifying them.

For how the certificate side of the system fails when DNS is taken over, see this site's piece on ccTLD hijacks and fraudulent certificates. That incident and this one are opposites: there the certificates were the problem; here they were the only part that was fine.

//What Defenders Can Take From It

Treat a network move as a security change. DigiCert now requires security sign-off and inventory updates for network access changes. An internal application's assumptions about who can reach it are invisible in its code.
An unknown account is a data question first. The triage question is not "how did it get created" but "what has it done". Blocking creation closes the door behind whoever is already inside.
Test the authorisation model, not the description of it. "No application assigned means no access" was true of the design and false of the endpoints. Per-endpoint checks were added only in the 30 October least-privilege review.
Move the alerting with the system. Account creation, permission changes and request volume on anything holding identity documents should reach the SOC, wherever the system sits.

[IOC]
Organisation: DigiCert (certificate authority); affected system: internal business system archiving validation data, separate from the CA, CertCentral and DigiCert ONE
Exposed to the internet: from 4 February 2025 (network migration) until endpoints were restricted to VPN on 9 July 2026 and moved to private addresses on 24 October 2026
Confirmed data access: 27 June to 2 July 2026, read-only, by a self-registered account without MFA
Data: names, business email, phone, business address; for a small minority, a passport, driving licence or utility bill. Count not published
Detection: unknown accounts seen 18 June; external report 21 June; data access suspected 12 October; reportable event 19 October; Bugzilla 2084460 filed 20 October
Certificates: none issued, modified or revoked; none need replacing

//Sourcing and Uncertainty

The timeline, the three contributing factors, the data categories, the 401 fix, the WAF and VPN restrictions and the thirteen action items are from DigiCert's report in Bugzilla 2084460, read on 7 November 2026. The two quotes about read-only access and the "account-creation issue" framing are from BleepingComputer's 5 November report, which also notes that DigiCert did not answer how many people are affected. The report's action-item table still showed items 12 and 13 as ongoing when read, while later comments say both were completed on 6 November. The "nine days" reading and the phishing assessment are ours.

What is unknown. How many customers and individuals are affected; what triggered the 12 October suspicion; whether the system was accessed before June 2026 and whether logs exist to say; who created the accounts.