There is a rule in Italian journalism that predates chat interfaces, is well written, and gets applied every day: you publish what is needed to understand the fact, not everything you know. It is worth picking up again, because it is already the answer to a question that looks new.
The test, as written
The rules of conductopens in a new tab on processing personal data in journalism (adopted by the Italian data protection authority on 29 November 2018) devote Article 6 to essentiality of information. Disclosing news of significant public or social interest stays within the exercise of the right to report where the information, however detailed, is indispensable because of the originality of the fact or of its description.
It is an elegant test because it makes no lists. It does not say “do not publish the address”: it says the address is published if without it the fact makes no sense, and not published if it only tells you where that person lives. The same set of rules returns to the point in the articles on the dignity of people who are ill, on the sexual sphere and on court reporting, which are the three places where the temptation of one more detail is strongest.
The test works, and it is applied seriously in newsrooms that take it seriously. The point of this piece is that it is applied at one moment only, and that moment is no longer the only one that counts.
Two sets, and nobody ever looks at the larger one
Take a real investigation. A four-hundred-page file, or three thousand emails, or two years of board minutes. Now consider two sets of people.
The first is who ends up in the story: four names, perhaps six. For each of them the newsroom will run exactly the Article 6 reasoning, with the title’s lawyer if needed. That set is small and watched.
The second is who sits in the document: the consultant who signed a report in 2019, the clerk copied into an internal email, the supplier on a minor contract, the family member named in an annex, the employee who filed it. Dozens of people with no part in the story, who will never appear on the page, and about whom nobody will ever run the essentiality test, because there is no reason to: they are not being published.
Then the whole document is uploaded to a chat to find the contradictions, and the second set (the large one, the one never examined) is transmitted in full to a third-party company. Essentiality was applied on the way out and never on the way in. It is the same test: only its position changes.
What the journalistic derogation actually covers
Here comes the objection, and it is a serious one: journalism has its own regime. Article 85 GDPRopens in a new tab requires member states to reconcile data protection with freedom of expression and information, and the Italian privacy codeopens in a new tab does so at Articles 136 and following: processing for journalistic purposes is allowed without the data subject’s consent, provided the rules of conduct are respected.
It is worth looking at exactly what that derogation removes. It removes consent, loosens the notice requirement and limits some of the data subject’s rights (objection, in some cases erasure) because otherwise anyone under investigation could stop it by asking for their own data.
It does not touch the security of the processing, and it does not touch recipients. The derogation exists so you can process that data without asking permission from the person you are investigating; it does not exist so you can transmit it to any supplier without thinking. Those are two different questions, and the second has no special answer for journalists: it has the answer everyone else gets, which is measures appropriate to the risk, and minimisation.
The same question, asked earlier
The practical advantage of applying essentiality on the way in is that the answer is almost always sharper than on the way out. At publication the test is hard: does that name help the reader understand the fact? Sometimes yes, and it takes a discussion.
At upload, no. The question becomes: does the model need to know what these people are called in order to do what I am asking? The real requests (summarise these minutes, find where the 2019 version contradicts the 2021 one, list the dates in order, translate the annex) are requests about the structure of a text. None of them improves because the model reads a surname instead of “[FULLNAME_7]”.
It is the same reason placeholders carry the work where a black rectangle does not: deleting breaks the sentence and often removes nothing from the file, substituting leaves the text usable and keeps the map to the real values on your computer, where you can delete it when the investigation closes.
A note on the archive
The rules of conduct also have an article on editorial databases and on protecting journalists’ personal archives, and it is not a period detail. A newsroom archive is a paper’s long memory, and investigations sit in it for years.
The question worth asking is not whether the archive is protected (usually it is) but how many full copies of that file exist outside the archive: in the desk’s chat, in the shared drive, in three people’s mail, in the history of an external service. The newsroom governs the first four. It does not govern the last, and that is the only one whose expiry date you did not write.