Incident analysis
8.8 million records and an invoice: the access nobody was watching
Denmark's population register was not broken into. A company's legitimate access to it was used, about fourteen million times, and the first thing to notice was the bill.
Access that is always on, has no limit on how much it can do, and is watched by nobody is a breach waiting for a password. The Danish case is unusual only in its size. The shape (legitimate access, used in a way nobody expected, found late) is the most common one there is.
What happened
On Monday 5 October 2026 the Danish government disclosed that names, addresses and CPR numbers, the personal identity numbers every resident is given, had been taken for about 8.8 million people. The figure includes people who have died or emigrated. Residents with name and address protection were not affected.
Nobody broke into the register. According to the CPR office, access was gained
by abusing a Danish company's legal access to search for information in the CPR
system
. Danish law lets private companies with a legitimate interest look up people
they have already identified, and one small IT company held that access. Over roughly ten
days in September it was used for about fourteen million lookups.
It was found by accounting, not by monitoring. A ministry official confirmed it was
during invoicing on Friday evening
that a very large amount was billed, and that bill
led the CPR office to the activity. The company's access was revoked, the police took the
case, and the minister acknowledged that security around the register had not been good
enough.
What is claimed, not confirmed
Three days after the disclosure, someone claiming to be responsible told the newspaper
Politiken that they had used a former employee's login, and that its password was
123456. Politiken later reported that several of the company's accounts,
an administrator's among them, used the same password. The authorities have not
confirmed any of it, and it has not been established whether the former employee worked
for that company. The rest of this post does not depend on it being true.
Three things went wrong, and none of them was the register
1. The access had no shape
The company was allowed to look people up. Nothing about that permission said how many, when or for how long. A small firm held access that could, and did, read most of a country. When access is granted as a yes or no, the only limit on misuse is how fast the attacker's script runs.
2. It was always on
Lookup access granted for a business purpose sat there between uses, every night and every weekend, waiting for whoever held a working password. If the claim about a former employee is true, that password had outlived the job it was given for. Even if it is not true, the access itself outlived any particular need for it.
3. The record existed, and nobody read it as security
Every lookup was counted, because every lookup was billed. The signal was in the system from the first day. It reached a person only when finance did its monthly job. The data that would have caught this in hours was being kept for a different reason.
Five questions to ask about your own access
You probably do not run a national register. You almost certainly run something with an export button, an API key or a reporting account. Ask these about it this week:
- Which accounts could read everything tomorrow? Not who is supposed to, who can: reporting users, integration accounts, support tools, the analytics connector. List them by what they reach, not by who owns them.
- Which of them belong to someone who has left? Compare that list with your leavers from the last year. An account that still works for a person who does not is the cheapest breach there is.
- Which of them are on when nobody is using them? Access that is needed for a monthly task does not need to exist for the other twenty-nine days. Make it something people request for the day and lose by the evening.
- What would a thousand times normal look like, and who would see it? Pick the number (lookups, exports, rows read) and send it somewhere a person reads daily. If the answer is "the invoice", you have the Danish setup.
- Can you say who held it on a given night? When something does go wrong, the first question is who could reach it then, not now. If that answer has to be pieced together from logs, it will be pieced together late. We wrote up how to answer "who has access to what".
Where time-bound access helps, and where it does not
TemprBac would not have sat in front of the Danish register. It grants people membership in the groups that carry access in your own systems (Entra ID, Okta, Google Workspace, GitHub, AWS IAM and Active Directory) for a window of one to eight hours, and removes it automatically. Read against this case, that changes two of the three things above:
- Access that is not on cannot be borrowed. When the group that carries an export or a lookup right is empty by default, a leaver's password reaches nothing. To get anything, someone has to ask, give a reason, and usually wait for another person to agree, at whatever hour they try it.
- The record is the access. Each grant is a row: who asked, why, who approved, when it arrived and when it was removed. That is a record somebody can read the same day, not a bill that arrives at the end of the month.
It does not fix the first problem. A grant still says who and until when, not how many. Rate limits and alerts on volume belong to the system that holds the data, and a strong password policy belongs to whatever people sign in with. Time-bound access shrinks the window. It does not count the lookups made through it.
Sources
- CPR breach: misuse of trusted access, CSIS, October 2026
- Hacker claims password '123456' enabled access to Denmark's national registry, IBTimes UK, 15 October 2026
- Denmark's CPR breach: securing third-party access to your API, WorkOS, 14 October 2026