How should a state prepare its population for artificial intelligence (AI)? Malta is trying to answer this question. Anyone who completes a basic AI course there should receive free access to ChatGPT Plus for one year. I think the approach is too short-sighted. A commentary analysis.
How should a state prepare its population for artificial intelligence? In May 2026, Malta gave a remarkable answer together with OpenAI: Anyone who completes a basic AI course in Malta will then receive free access to ChatGPT Plus for a year.
I think this approach is exciting, but not sufficient, especially not transferable to Germany or other countries. Because free access to a single AI model is too short-sighted; Rather, what is needed is a broader empowerment strategy.
ChatGPT subscription: What’s behind Malta’s AI deal with OpenAI?
In May 2026, Malta announced a program together with OpenAI that is attracting international attention: citizens and residents of Malta will receive free access to ChatGPT Plus for one year after completing an AI course.
OpenAI itself speaks of the world’s first partnership of this kind. It is part of the Maltese initiative “AI for All”, which is intended to introduce people of different age groups to the responsible use of AI.
The idea behind it is simple: Malta doesn’t just want to regulate AI or leave it to companies, but rather bring AI to the broader society according to the motto “First learn, then use.”
Good idea, weak implementation: what Malta is doing right and wrong
I have to say right away that I think the idea behind the Maltese approach is good. In my experience, training people alone is of little use if what they have learned cannot be applied. Access to ChatGPT Plus for one year at least gives training participants the opportunity to use it for this period.
On the other hand, I don’t think the implementation of the idea, as Malta has done, is that good. Other countries, such as Germany, could act more cleverly here. What do I mean by that?
We in Germany should at least take a look at the Maltese model and think about something comparable. Article 4 of the European AI Regulation already obliges providers and operators of AI systems to do their best to ensure a sufficient level of AI competence among people who handle AI systems on their behalf.
Why “ChatGPT for everyone” would be the wrong signal in Germany
This does not result in a general obligation for a state, including the German one, to offer AI training to all citizens. But it shows a clear direction: AI use without competence will no longer be accepted as the norm, based on the general understanding that the population needs to be “smartened up”.
Nevertheless, “ChatGPT Plus for everyone” would be the wrong signal for Germany. Not because ChatGPT isn’t a powerful tool. But because a state has to be careful when it de facto declares a certain private platform to be the standard interface for AI. Then AI is quickly equated with chatbot, text generator, US platform and subscription model. But that would be far too restrictive.
AI competence: The problem is not just access
There is also something else: Many people have long had access to AI: via search engines, office applications, smartphones, apps, browsers or business software. Access to AI applications is probably not the central problem. The real bottleneck lies elsewhere.
In my experience, many potential users don’t even know what they can use AI for. In addition, AI results are trusted far too quickly. And then there are users who generally reject AI because they have seen bad examples.
Companies, on the other hand, have additional issues that they have to address (which of course also apply to private users). For example, you ask yourself what is generally permitted, what data may be used and how employees must be trained without turning these training courses into an “alibi event” to fulfill the requirements of the AI regulation.
A free premium subscription to ChatGPT does not automatically solve these problems. In the worst case, it even makes it worse: people produce texts, analyzes or decisions more quickly, but without being able to better assess whether these results are correct, legally clean or even usable.
Three criteria for a better AI strategy
Rather, I believe we should think in the direction of how we enable people to use AI sensibly and beneficially. So it’s about an enablement strategy. For this I see three necessary criteria that must be met.
Firstly, it would need an offer that is provider-neutral. The state should not privilege any single platform, but rather promote open standards for training content, safe learning environments and documentable proof of competence.
Secondly, an offer would have to be target group-specific. An “AI for everyone” course sounds democratic, but it doesn’t make sense in terms of applications. Because a pensioner needs different application examples than employees in a craft business, and they need different examples than schoolchildren and students.
Thirdly, the offer would have to be practical. People learn AI not through abstract definitions, but through applications.
The better approach: An AI flight simulator for Germany
If I had to design a German approach (which I would really enjoy), I wouldn’t start with simple, time-limited access to AI applications, but with a protected learning space.
Incidentally, these are nothing new, but are known as “regulatory sandboxes” or “real-world laboratories”, which refer to controlled environments in which, for example, companies can test new ideas, products or services without being immediately confronted with the full force of regulatory requirements.
In the present case of AI, one could – to use a new term to highlight the AI situation – speak of a public “AI flight simulator”.
There, citizens, companies, schools and administrations could practice the use of AI with realistic but fictitious examples, for example recognizing fake news, checking AI answers, identifying personal data, improving bad or weak prompts or making an administrative message more understandable.
The advantage of such an AI flight simulator would be that humans could make mistakes without causing real damage. And the “wheel” doesn’t even have to be reinvented, because comparable systems already exist, for example for teachers in Bavaria, at least as far as they are provided by the state.
Why mistakes are the best AI course
AI competence would arise through experience and not through abstract warnings or training. Because people need to experience how plausible wrong answers can sound.
You have to see how easily confidential information gets into the wrong systems. And they have to learn that better questions produce better answers – but they don’t replace an exam.
Such an AI flight simulator would have to be available online, but not only that. Access to online offers in particular is not a given in Germany.
Access would therefore have to be made possible via other channels that already exist here, namely adult education centers, libraries, chambers, schools, universities, job centers, senior citizens’ organizations and corporate networks. Using such an AI simulator, corresponding basic AI skills could then be taught and certified.
The crucial question is not access, but ability
Malta has opened an important debate regarding AI. AI is no longer just a topic for companies, developers or ministries. It’s about population, participation, education, work, consumer protection and democratic resilience.
However, I don’t think simply providing free access to an AI tool is sufficient because it doesn’t involve a competency strategy.
Rather, we should learn from the Maltese approach and develop it further. It would make sense to have an approach that trains people in real situations: checking information, recognizing manipulated content, using AI at work.
The question for the future is therefore not whether people will have access to AI. They get it anyway, even if it is via a free entry-level model available online.
The crucial question is how to enable them to classify, check and use AI results responsibly. I believe that the answer to this question will show whether AI only makes things faster – or enables useful applications.
Also interesting:
