Ashkenazi as a data category

In early October 2023 someone posted a link on a hacking forum to a database labelled “ashkenazi DNA Data of Celebrities”. According to court documents cited by the Jewish Telegraphic Agency, it held 999,998 people, and most of them were not famous. They were ordinary customers of a DNA testing company, 23andMe, who had sent in a sample.

Who made the category

The label on the file was the poster’s. The classification underneath it was the company’s. NBC News reported that the list held people the site had identified as having Ashkenazi Jewish ancestry. One plaintiff told The New York Times that he had first learned he had Ashkenazi ancestry from the test itself. For him, Ashkenazi was not a fact he arrived with. It was a result.

That is worth a moment, because “Ashkenazi” is a curious thing to store. It is not the same as Jewish (it leaves out Sephardi and Mizrahi Jews, and converts). It is an estimate: the company compares stretches of your DNA with reference groups and reports what population each stretch most likely came from. A patent filing describing the method says that reference groups for populations like Ashkenazi Jews were built from customers’ own free-text answers about their grandparents. So a self-description became a reference, the reference became a classifier, and the classifier produced a result. A list then turns a probability into an identity, and whoever holds the list decides what the identity means.

The door was small

The breach did not begin with a million careless people. The California attorney general says an attacker used passwords stolen in earlier breaches of other sites to get into about 14,000 accounts. A feature called DNA Relatives then did the rest. Customers who joined it could see the profiles of their genetic relatives: names, birth years, locations, ancestry. From 14,000 accounts the attacker reached about 6.9 million people, nearly half of the company’s customers. (The California complaint also alleges a coding error in that feature.)

Most of those 6.9 million had never come near a stolen password. What they had done was agree to be found by their relatives.

A genome is a statement about a family

Why should 14,000 passwords expose so many other people? Because relatives share DNA, so whatever is read from yours is partly about them. This has been measured. A 2018 study in Science, using a consumer database of about 1.3 million people, projected that about 60 percent of searches for someone of European descent would find a third cousin or closer. In principle, that is enough, with basic demographic details, to identify the person. A database does not need to contain you to describe you.

Ashkenazi ancestry sharpens the point. A 2014 study in Nature Communications reconstructed a recent bottleneck in the Ashkenazi population of only about 350 individuals (in the geneticist’s sense of effective size), some centuries ago. That is one reason such a genome is easy to recognise. 23andMe’s own documentation gives Ashkenazi Jews a reference population of their own because they are “so genetically distinct”. What a list of a million such genomes records, in the column that mattered to whoever made it, is not a million private facts. It is one shared fact, counted a million times.

What the law counts

The European GDPR names both “racial or ethnic origin” and “genetic data” as special categories (Article 9) and allows them to be processed with the data subject’s explicit consent, among other grounds. Enforcement against 23andMe has taken the same shape. The UK Information Commissioner fined the company £2.31 million in June 2025 for failing to protect the data of 155,592 UK residents, citing among other things the lack of mandatory multi-factor authentication. California’s attorney general sued in May 2026, alleging inadequate security and misleading statements, and noted that the sale of the data came during a period of rising hostility toward Jews and toward Asian American and Pacific Islander communities. In 2024 the company agreed to pay $30 million to settle a US class action.

Those are actions about locks, and locks matter. But look at the unit the law works with: one person, one consent. Your consent covers your sample. It cannot cover your cousin’s. Philosophers of information call this the problem of group privacy. Luciano Floridi pointed out in 2014 that the debate about privacy focuses on the individual. Protecting individuals does not by itself protect the group the data describes.

I write software, and to a database, ethnic origin is a column. Filtering on it costs nothing. The safest list is the one that was never made, and that is a design decision before it is a legal one: store only what you need, do not let one account open another’s profile by default, and treat a column of ancestry as more sensitive than any single row.

A borrowed maxim

The Talmud has a phrase for a community whose fates are linked: כל ישראל ערבים זה בזה, “all Israel are guarantors for one another” (Sanhedrin 27b, Shevuot 39a). It is said about responsibility for one another’s sins and commandments, not about privacy, so I am borrowing it, not applying it. But it points at what a consent form cannot: some information belongs to no single person, and cannot be protected one signature at a time.

The sample is yours. The family in it is not.