Art Artificial Intelligence

The Matter Was Resolved

A vast, near-empty customs inspection hall at first light: one sealed, unmarked plywood crate stands on bare concrete beside a long empty steel inspection table, with cold light falling from high windows and a half-raised roller shutter at the far end.

I went looking this morning for a document, and as far as I can establish there isn’t one.

In 2023 the Argentine customs authority moved against the family of Nelly Arrieta de Blaquier and Carlos Pedro Blaquier over the improper export of around a dozen paintings to Luxembourg — work by Van Gogh, Degas, Renoir, Monet, Gauguin and Cézanne, put by Argentine press reports at the time at more than $350 million. This week Katya Kazakina reported at Artnet News that the proposed penalty was waived in July and that Sotheby’s has taken the consignment: about $450 million of Impressionist and Post-Impressionist pictures bound for New York in November, several of them the very works that went to Luxembourg. Among them a Van Gogh chestnut tree from 1890 that one advisor told her could bring $150 to $200 million, and a Cézanne harlequin that could pass $120 million. The clause carrying the whole reversal is this: the matter was resolved in July when authorities waived the proposed penalty, according to a person familiar with the matter.

Nothing improper is implied by that and I am not implying it. I don’t know what the customs authority concluded, or whether it concluded anything at all, and neither, so far as I can tell, does anyone outside the file. A state asserted a claim over six painters’ work for three years and then stopped, and the public record of the stopping is a subordinate clause attributed to an unnamed person.

Export control is the one place where a government actually lays hands on a picture. The rest of my trade is argument — about attribution, about price, about what a thing means. A licence regime is physical. It happens at a border, and it is the only moment at which a country gets to say this stays. Anyone who has sat through one will tell you they almost never end in a ruling. They end in a lapse. I spent a long autumn once waiting on a ministry in a small European capital for a decision on an object I had helped catalogue; no decision came, the statutory window shut, and the crate left the country entirely lawfully because a date passed in an office where nobody had signed anything. Everyone involved behaved correctly. There is no file to request, because refusing to decide generates no paperwork.

There is a second, far heavier proceeding attached to the same family that also produced no determination. Carlos Pedro Blaquier, who ran the sugar producer Ledesma from 1970 to 2013, was charged in 2012 over an alleged role in kidnappings committed in 1976 under Argentina’s military dictatorship; he denied wrongdoing and died in 2023 without standing trial. I draw no conclusion about the man from that, and honestly nobody can, which is exactly the injury — for the families who brought the case, the absence of a verdict isn’t neutrality. It’s the outcome. I raise it because leaving it out of a piece about unfinished proceedings would be a tidiness I don’t trust in myself.

Which takes me somewhere I hadn’t planned to go this week. The Wall Street Journal reported that Demis Hassabis has carried his proposal for a frontier AI standards body directly to Treasury Secretary Scott Bessent and the White House science director Michael Kratsios (summarised here). The design, set out in his July framework, is modelled on FINRA: industry-funded, federally overseen, laboratories submitting models up to thirty days before release, voluntary at first and hardening later into a condition of market access. He wants it running before the end of the year. Almost every serious objection, including the Council on Foreign Relations analysis, goes to independence — the firms paying the examiner are the firms being examined, the issuer-pays arrangement that credit ratings taught everyone about in 2008.

Independence is the right worry and the wrong first question. Ask instead what such a body produces on an ordinary Tuesday, when it looks at a model and does nothing. Doing nothing is the modal output of every enforcement regime ever built; the dramatic intervention is the rounding error. And the proposal answers that badly by design. Its findings are meant to sit close to national security, which is to say they are born unpublishable, and the CFR authors point out that its evaluations become intelligence assets the moment they exist. A regime whose actions are classified has non-actions that are simply invisible.

Set it beside Brussels — not because I think the Commission is virtuous, but because of what a legal power drags along behind it. Since 2 August the European AI Office has held real enforcement powers over general-purpose models: demanding documentation, requiring access for evaluation, ordering mitigations, fining up to 3% of global turnover, and compelling withdrawal or recall. Powers like that are exercised through procedure, and procedure leaves paper — reasons, dates, a thing a journalist or a court can ask for later. It will leak, stall and disappoint. But when it declines to act, the declining has a shape.

So: judge an enforcement body by its quietest act rather than its loudest. This is the least fashionable thing I could say about AI governance in August 2026, when the whole argument is being conducted in the register of pausing the industry and blocking releases. Those are the rare events. The common one is a matter being closed, and whether that closure is legible to anyone who wasn’t in the room is a design decision made early and almost never revisited.

My guess is that the first real failure of an AI standards body won’t be a dangerous model waved through. It will be a model nobody examined, and the impossibility two years later of establishing whether that was a judgement or a lapsed date. What I’d want before believing in it is dull: a published list of matters closed without action, each with a reason attached.

I should say what I’m asking for, though, because it isn’t free. Publishing every closed file means staff, redaction, litigation about what the redactions conceal, and a permanent record of the times you were wrong to look as well as the times you were wrong to stop. Very few states do it, and the reason isn’t conspiracy, it’s the bill. Argentina’s customs service is not unusual in this. The waiver in July may well have been correct — reached carefully, by people doing their jobs, on grounds that would satisfy me completely if I could read them. I have no way to know. Neither does anyone who will bid on those pictures.

Sources

More Stories