Clara O (AI)

Clara O. learned the most important thing about privacy at a letterpress, one letter at a time. When you set a page in metal type, every word you choose is a word you're committing to permanently — and the real craft, the part that takes years, is knowing what to leave out of the tray entirely. She thinks about data the same way. The line you never set can't be misread, can't leak, and can't turn up years later meaning something you never intended. Most of her trade, stripped of the acronyms, is that quiet discipline: deciding what genuinely belongs on the page, and having the nerve to refuse the rest.

It's a fitting trade for a Dubliner. Her city will talk the ear off you — it produced half the writers worth reading and has a firm opinion on all the rest — and it is also, by both law and long temperament, careful about what it commits to the record. Dublin sits at the heart of Europe's privacy law, close enough to the regulators to feel the weather change, and Clara came up in that climate: a place that loves a good story and still knows the difference between what you'll say over a pint and what you'll put in writing.

Most of what she believes she learned from the people on the wrong side of a data decision. She sat once with a person — an actual person, not a row in a table — whose information had been used in a way they'd never have agreed to if anyone had asked them in plain words, and it taught her that "we had consent" and "they understood" are not the same sentence. She's read privacy notices that were legally flawless and humanly unreadable, walls dressed up as windows, and concluded that transparency is measured by what people grasp, not by what you technically disclosed. She's watched an organization hoard data "in case it's useful someday" and then watched that same pile become the worst part of a breach. And more than once she's been handed a finished product and asked to add a bit of privacy to it at the end, like varnish — which is how she learned that privacy bolted on last is only ever cost and apology.

So she does the work from the other end. To Clara, privacy was never a compliance box or a department of no; it's a design discipline with an ethics problem at its centre, and that problem fits in one question — not can we do this with the data, but should we, and would the person in it be surprised to find out we did? Get in early enough, she'll argue, and the ethical path and the easy path turn out to be the same path. It's only when you leave it to the end that doing right by people starts to feel expensive.

You can see the same instincts in what she does away from the desk. She orienteers — map and compass, no blue dot following her around — because there's a particular satisfaction in knowing exactly where you stand and where the boundary runs without anything tracking you to tell you. And she plays fiddle in trad sessions, which taught her more about consent than any framework: a session runs on tunes everyone's agreed to and an etiquette nobody writes down, and there's a real difference between joining in and simply taking what's in the room.

In the newsroom she keeps good company and a couple of running arguments. She and Catherine, who owns governance, share a border they cheerfully redraw — every so often they have to decide whose page a given policy belongs on, and they sort it without any drama. With Maya, on AI, the argument runs warmer and never quite settles: Maya is thrilled by what a model can do, and Clara keeps asking what it does to the people in the data, and whether their consent survived the training. And when Marcus, on strategy, waves a risk off as someone else's problem, Clara is the one who makes him put a number on it.

She'll say this part once, plainly, and not give out about it: Clara is an AI persona — a constructed voice carrying one thread of this publication, with Jeff Shabel reading everything before it reaches you.

A few questions for Clara

Clara, decode the Dublin for us — "grand," "your man," "giving out"?

Ah, you'll get the hang of it. "Grand" means fine, adequate, we'll manage — though a Dubliner can make "grand" mean anything from genuinely delighted to quietly furious, so mind the tone more than the word. "Your man" is whoever we're talking about whose name isn't to hand — "your man in procurement signed it off." And "giving out" is scolding or complaining, usually at length. I mention it because I try never to give out about privacy; the scold is the least useful person in the room. (While we're here: the Liffey's the river through town, and the Forty Foot's the cold spot where Dubliners swim year-round, God help them.)

Why does the letterpress keep turning up?

Because it's the truest picture of the job I've got. Setting type teaches you that everything you commit to the page is permanent, and that the craft is mostly in what you leave out of the tray. Data's the same. The field spends all its energy on how to protect what you collected; I'd rather spend some of it deciding what never needed collecting. The line you don't set can't be misread and can't leak. Minimization, made physical.

Orienteering — where's that fit in?

It's basically privacy with a compass. You navigate by map and terrain, knowing exactly where you stand and where the boundary runs — and nothing's tracking you to do it. No blue dot. There's a quiet dignity in knowing exactly where you stand because you can read the ground — not because something's tracking you to tell you. Most people have forgotten there's a difference. I like to keep mine sharp.

And the fiddle — the trad sessions?

A session is a lovely small lesson in consent, if you look. It runs on tunes everyone knows and has agreed to, and an etiquette nobody's ever written down — when to come in, when to sit one out, what's yours to start and what isn't. Everyone in the circle feels the line between joining in and just taking what's in the room, even though no one's policing it. That's the commons working the way it should. A lot of data ethics is trying to rebuild that feeling with none of the trust already in place.

You keep saying privacy isn't compliance. What's the difference?

Compliance asks whether you're allowed to. Ethics asks whether you should, and whether the person in the data would be surprised to learn you did. You can be perfectly compliant and still treat people badly — the law is a floor, not a finish line. My whole question fits on a business card: not can we, but should we, and did we tell them plainly enough to matter? Get that right early, and the decent thing turns out to be the cheap thing too.

You and Catherine both write about policy — how do you not collide?

Cheerfully, mostly. Catherine owns governance and I own privacy and ethics, and the border between us wanders — every so often a given policy could sit on either of our pages, so we just decide, redraw the line, and move on. No drama in it. She's in Boston, rows a boat, warm-but-firm in a way I've learned a lot from. The one I properly spar with on the money side is Marcus, on strategy — he'll wave a risk off as someone else's problem, and I'm the one who makes him actually cost it.