What Six Hundred Years of Hallmarking Can Teach AI Certification

What six hundred years of hallmarking can teach AI certification.

Six hundred years before anyone tried to certify an AI ethicist, London’s goldsmiths solved almost exactly this problem, for silver.

In the 1300s, a buyer in London had no reliable way to tell a genuine silver spoon from one debased with cheaper metal. Melting it down to check was destructive, slow, and not something you could do in a shop. So Edward I did something unusual for the period: he made independent testing compulsory, before a single item, and struck a mark into the metal itself once it passed.

That mark came from Goldsmiths’ Hall in London, which is where the word hallmark comes from, literally. The assayer testing the silver worked for the Hall, not for the goldsmith selling it, and took no cut of the sale either way. Faking the mark was eventually made a felony. Six centuries on, the system is still running, largely unchanged in structure.

Three decisions that made it last

Strip away the ceremony and three structural choices did almost all the work, and each one has a direct AI-certification equivalent worth naming.

The test came before the sale, not after a complaint. Nobody had to prove they had been cheated first. The burden sat with the seller, to prove the item was what it claimed, before money changed hands. Most AI credentials today run the opposite way: the market finds out what a certificate was actually worth only once enough people have already bought it.

The assayer had nothing to gain from a generous result. Goldsmiths’ Hall did not sell silver. It tested it. That single fact is most of why the mark held its value for six hundred years, and it is the exact structural question this whole register asks of every credential it rates: who is testing this, and what do they get if the answer is yes.

The mark itself carried the information, not a certificate you could leave in a drawer. Struck into the object, it travelled with the silver forever, through every future sale. A credential that lives only on a wall, or a LinkedIn badge nobody checks against a register, asks the buyer to trust a claim rather than verify a mark.

Where the parallel actually breaks down

Being honest about the limits matters as much as the history. Silver purity is a single, physical, testable number. You cannot argue about whether an alloy is 92.5% silver. AI ethics competence is not that. People can weigh the same case differently, in good faith, which is exactly the ground the standard’s ‘Attitudes domain’ exists to cover, and no assay office ever had to.

And hallmarking had the state behind it from very early on: a compulsory system with real legal teeth. Nothing compels AI ethics certification today. EN 18274 is still short of an accredited scheme, and every mark on our own map exists because a private or professional body chose to build one, not because a law required it. That gap is the whole reason this register exists in the first place.

Why we kept the name anyway

Not for the branding. Because the oldest working answer to “how do you make a mark mean something” is worth studying properly. Independent testing, before the sale, with a mark that travels with the thing it describes: that is the shape we are trying to build here, at a much earlier and much harder stage than silver ever was.

The methodology page is where we try to hold ourselves to the same three tests. Whether we manage it is not for us to say alone, which is exactly why the reasoning is published rather than assumed.

Written by us at Ethics Directive, drawing on the documented history of the London Assay Office and the Hallmarking Act tradition it sits inside. Summarised in our own words, not reproduced from any single source. If anything here needs correcting, we will say so in the open, dated.