Despite common belief, the Ottomans never banned the printing press. They did something more effective: made it a matter of licensing.
In 1452 a Hungarian offered the Byzantines a cannon. The emperor Constantine XI could not afford him, but the Ottoman sultan Mehmed II hired him instead, and within a year Orban's guns were breaking the walls of Constantinople - marking the end of the Middle Ages. A couple centuries later, in 1727, another Hungarian-born convert, İbrahim Müteferrika, wanted to give the Sultan a printing press. He was made to write a treatise arguing for its usefulness before he got a licence, and in thirteen years his press produced only seventeen titles. Europe, by 1500, already had presses in 282 cities producing millions of books.
The popular version of this story blames scribes protecting their monopoly. Luddism 200 years before Ned Ludd. However, historians have found no evidence of the absolute ban. What existed was a licensing system, and the best economic account of it (by Coşgel, Miceli and Rubin) explains its purpose. Mass printing threatened the religious authority the sultans relied on. The same empire adopted military technology (which reinforced rather than threatened authority) as fast as it could pay for it.
That is the lesson worth keeping: The Ottomans did not fear new machines for the sake of it. Technologies that concentrated power at the centre were hired on the spot, whereas those that distributed it were seen as a threat and deliberately slowed down through a licence.
I believe Europe is running the same play on AI.
In some aspects, the EU is already running at cannon speed (as it should). In July the EU opened its tender for up to seven AI gigafactories, with up to €10 billion in public money meant to draw more than €20 billion from private investors.
However, when it comes to user adoption, it’s more like the Ottoman printing press, probably an unintended consequence. In 2025, 55% of large EU firms used AI, against 17% of small ones. The technology is reaching the organisations that already concentrate capacity and stalling at the ones that would spread it.
AI legislation and regulation are not the main reasons small firms hold back. Among European firms that considered AI and passed, roughly seven in ten cite a lack of expertise, and just over half cite unclear legal consequences. However, for a small firm those two are functionally the same barrier. A large company treats legal uncertainty as a line item: it hires counsel and moves on. A twenty-person firm can't, because the expertise it lacks is precisely how to read the rule and take reasonable risks within it. A permission regime doesn't need to forbid anything to effectively slow things down, it just needs to make adoption scary or require a specialist that small actors can't afford.
The AI Act's high-risk obligations cover recruitment, credit scoring, education and essential services, which are exactly the places where AI enters ordinary businesses. They have been in force since 2 August 2026. The harmonised standards and national authorities weren't ready, so the Digital Omnibus entered into force six days before the deadline and moved it to December 2027. Classic EU move. The national regulatory sandboxes, the one tool designed to let small firms experiment cheaply, slipped by a full year.
Move the scrutiny to where power concentrates
If permission taxes the actors who diffuse a technology, the fix is to put scrutiny where capacity concentrates.
Obligations should scale with the deployer, not the category. A bank scoring a million borrowers should carry the full weight of conformity assessment. A fifty-person staffing agency using an off-the-shelf screening tool shouldn't, and doesn't need to: discrimination law already applies to its hiring, whatever tool it uses. A small deployer using a compliant model for its stated purpose should get a presumption of conformity, putting the legal burden on the provider, who can afford the lawyers. And no obligation should apply before its standard exists. If the standards aren't ready, the date moves automatically.
That keeps the guardrails where the harm scales and makes going first cheap for everyone else. We have done something similar with GDPR to great success in the past decade, achieving both the most comprehensive data protection regulation in the world, and one that barely slows down companies looking to get off the ground. Its teeth bite harder the larger the offender and the more egregious the violation; but honest actions in good faith are rarely punished. Great cost/benefit balance.
The Janissaries
As a teenager, I loved playing Age of Empires II with my friends. I often picked the Turks, the gunpowder civilisation, and their Janissaries could flatten a line of pikemen in a few attacks.
The game got the Ottomans right for 1453 and wrong for 1700. The Janissaries began as the sharp end of the concentrating technology. Over the next two centuries they became a privileged corps with its own political interests, and Ottoman military technology stagnated while Europe kept innovating. The soldiers who carried the guns came to resist the reforms that would have modernised them, until Mahmud II destroyed the corps in 1826. The cannon-makers became the scribes.
Europe should build its gigafactories. But a continent that funds the concentrating technology at speed while routing everything else through a licence isn't avoiding the Ottoman mistake. It's repeating it, and raising its own Janissaries along the way.
Preview image: janissaries of the Ottoman army, Ottoman miniature. Public domain, via Wikimedia Commons.