Anthropic has written rules against game cloning. Ars Technica says they won’t hold.

On 2 November 2026, Ars Technica reported that Claude now has new rules against cloning games. The piece’s argument is that the rules won’t stop the AI copying boom, because the wave of imitators is far bigger than any one model maker can stem. That is the whole of the news, and the argument is the interesting part.

The hope is that a rulebook can hold the line

The excitement, or the relief, runs like this: if the big models refuse to clone, the copying slows down. I test that the way I test every claim. Is there a real need? Yes. Studios and software makers have spent money building things, and they want those things protected. What actually changed? One model, under one company’s policy, now declines a certain kind of request. What didn’t change? Every other model. The open models anyone can run on their own machine. And the plain fact that a clone never needed a model in the first place. A developer with a screenshot and a weekend could always rebuild a simple product. AI shortened the weekend to an afternoon.

A rule inside one tool changes what that tool does. It does nothing to what the market can do. So the protection on offer is thin, and I would not plan a business around it.

Vibe coding made this unavoidable

This is the vibe coding point I keep coming back to. Vibe coding collapses the distance between an idea and working software. That is the only distance it collapses. If someone can describe your product in a few paragraphs, they can now have a working version of it quickly, and no usage policy will change that for long.

Which means a product whose only asset was its code was always going to be copied. AI moved the date forward. It did not create the exposure.

What vibe coding doesn’t touch is what decides commercial success: real-world testing, feedback from actual users, marketing and customer acquisition. A clone gets the features on day one. It doesn’t get the customers who already use yours, the channel that brings new ones in each week, the partner who sends referrals, or the two years of support conversations that taught you what people actually struggle with. Those are built in the market, over time, and no prompt generates them.

What this looks like for an operator

Take a shop that paid for a custom ordering app, or a marketing team that built its own reporting tool with Claude. Suppose a competitor copies it exactly. What has the shop lost? Not the orders. The orders come from customers who know the shop, the WhatsApp list it has built, the reviews, the supplier who gives it first pick of stock. The app was a convenience layered on top of a business that already existed. Copy the app and you still have no business.

If you sell software, the same test runs in reverse. Ask what you own that a competent developer with Claude could not rebuild this month. If the honest answer is “the code”, you are already exposed, with or without Anthropic’s rules. If the answer is “four hundred paying customers, a referral channel and integrations that took a year of meetings”, you have something.

I see this in AeroChat. At the chat level, every AI chatbot is almost the same technically, and if AI is the only selling point, it is not a selling point at all. What is hard to copy is the integration with each market’s payment processors and logistics partners, which global vendors skip. You can’t vibe code your way into those relationships. They take introductions, contracts and time.

What a business should do

Stop treating a model maker’s rulebook as part of your defence. Assume you will be copied and plan from there.

If you buy or build software for your own business, stop paying a premium for exclusivity of code. It won’t stay exclusive. Pay instead for who keeps it working, who answers when it breaks, and how well it holds your own customer data and processes.

If you build and sell software, count what you own beyond the code. Customers, the list, the partners, the data from real use, the channel that brings people in. Spend the next quarter on those, because that is the part a clone can’t generate.

If you are deciding whether to build something at all, ask the vibe coding question before you start. Can I get this to customers, and will they tell me what to fix? If the answer is no, a cheaper build doesn’t help you.

Source: Ars Technica: Claude’s new rules against game cloning won’t stop the AI copying boom