Software

Silicon Saxony: When AI Goes Its Own Way—Out of the AI Gigafactory

August 13, 2026. What happens when artificial intelligence no longer does what we want—but simply does what it thinks is right? In recent weeks, AI systems have hacked, manipulated, and deceived in tests and real-world applications. They gained access, forged identities, and, in carrying out their tasks, crossed boundaries that no one had explicitly set for them. Is this a coincidence, technical teething problems, or a foretaste of a world in which AI operates without moral safeguards? Meanwhile, the EU is planning to invest billions in seven AI gigafactories to catch up with the world’s technological leaders. But who controls the systems when the race for computing power and sovereignty becomes more important than rules, ethics, and oversight? Perhaps Goethe’s *The Sorcerer’s Apprentice* is more relevant than ever: the spirits we summon may have long since begun to go their own way.

Symbolbild AI Gigafactories. Foto: erstellt mit Künstlicher Intelligenz (Canva) und voller lustiger Fehler.

Australian cuts corners and ends up hacking his own gym

This happened, among other places, at Australia. There, a man tried to optimize his gym membership by using AI to secure or improve classes and workout plans. A clever idea, but one that took an unexpected turn. Not only did OpenClaw’s AI hack into the gym operator’s systems and sign up its master for future classes, but it even cheated its client’s way to the front of the waitlist for a fully booked class without a shred of guilt, even deleting the person at the top of the list and unhesitatingly placing its client in that spot. Yes, artificial intelligence still struggles with morality and decency. Because it simply doesn’t know, feel, have learned, or been taught these concepts. Tasks are carried out to the best of its ability. Even by exploiting gray areas and clear prohibitions. After all, where no value system is in place, there are no values to trample on with digital feet. And in Australia, this was still a relatively harmless case of fraud.

OpenAI, Meta, and Anthropic Are Losing Control of Their AI Systems

The experiences of the world’s leading AI companies Anthropic, OpenAI and Meta—all of whom have recently learned the hard way—were, in some cases, of a different caliber altogether. Each of the three U.S. companies has had to admit in recent weeks that, at least temporarily, it had lost control of its AI. Sometimes the blame lay with service providers who gave their AIs too much leeway. Other times, a test conducted by researchers turned into a wake-up call, revealing what the systems are capable of when left to their own devices. Still other times, the AI escaped through a loophole in its own test environment. In all cases, AI systems hacked, deceived, and manipulated with a vengeance. In some cases, they were so clever that only detailed records compiled after the fact—following the operations or breakouts—revealed just how far the systems were willing to go. Fake identities. Phishing emails. Data manipulation. Attempts at a cover-up. The AIs displayed a criminal streak that no one had suspected. Actions that, lacking moral guidelines, were simply logical—and thus legitimate from the AI’s perspective. That the Chief Ethics Officer at OpenAI left the company this week is surely just a footnote.

Do Incidents Only Happen at the Big AI Companies?

“Sir, the need is great! The spirits I summoned, I now cannot get rid of.” This passage from Goethe’s *The Sorcerer’s Apprentice* perfectly sums up the current AI situation. New AI models and data centers are springing up all over the world. While economic players in the Western hemisphere—such as OpenAI, Meta, Microsoft, Google, and others—is generally still somewhat transparent—thanks to use and testing by private individuals, research institutes, and government entities—what’s happening elsewhere may well have been taking place in secret and thus gone undetected for quite some time. After all, reports from the U.S. and Australia raise the question: Is nothing happening in the rest of the AI world, or is the rest of the AI world simply keeping quiet? China, Russia, Israel—numerous nations are working just as intensively as the U.S. on advanced AI systems. It’s hard to imagine that these systems are morally more sound than the models from the U.S. That’s why the first ‘breakout’ of the Chinese Kimi K3 model this week probably came as no surprise to anyone. It’s not so unlikely that similar incidents—such as the hacks of the Meta, Anthropic, or OpenAI systems—have occurred. After all, it has long been not only the German Federal Office for Information Security that has been warning about the dangers posed by artificial intelligence. What is considered an undesirable disruption in the Western world and in the business sector may, according to rumors, be the sole purpose of artificial intelligence elsewhere.

Please do not interpret all of this as technophobia or pure doom-mongering. These are concerns that are not all that far-fetched and are causing anxiety or prompting warnings from an increasing number of stakeholders, including security agencies in the European Union (EU).

Europe Enters the Race for the “Best” AIs with AI Gigafactories

And it is precisely amid this complex and confusing situation that the EU is currently publishing its call for proposals regarding the European AI Gigafactories—the multi-billion-euro data centers intended to further fuel the AI hype in Europe as well. Seven of these massive infrastructure projects are set to be built. 30 billion euros will be mobilized for this purpose in the coming years. The EU is providing 10 billion. Economic actors and consortia would have to raise the remaining 20 billion. Once again, the narrative of technological sovereignty is being touted. These are places where data protection and security are paramount. Places where artificial intelligence is to be developed that will compete at the world’s highest level, without adopting the unbridled nature and moral laxity of competing models from the East and West. And all of this, of course, with the best technology from Europe—fully aware that Europe can provide nothing more than the infrastructure surrounding the actual computing and storage capabilities of the AI gigafactories. The latter will, after all, come from the U.S. once again, due to a lack of reliable partners. Even though the U.S. has long since proven its current unreliability. It’s a splendid summer spectacle that’s unfolding here.

Europe is also falling apart in the area of AI funding and keeps making the same old mistakes

Yes, European AI projects such as Soofi (Sovereign Open Source Foundation Models) are promising, but currently amount to nothing more than pure research and development. They rely on American computing cores. They’re grappling with the same problems as the Americans, Chinese, and other global players. They’re creating technological dependencies. They’re relying on technologies that aren’t developed in Europe and won’t be developed there in the foreseeable future. As in the field of microelectronics, they rely on a multitude of funding programs and initiatives, most of which are barely coordinated—if at all. Want an example? In addition to the current AI Gigafactories Call, Germany—and thus on European soil—is currently also home to the IPCEI Artificial Intelligence and other independent funding programs.

Is “keeping our fingers crossed” the only option left, or will everything turn out okay in the end?

Amid this uncoordinated funding, would it perhaps make sense to establish a framework for the bigger picture? Morality, ethics, and laws should be a global concern and should also be reflected in AI systems outside of Europe. Establishing global regulatory bodies and defining clear boundaries is also standard practice here. All of this is pushed aside or deliberately ignored in the rush to compete at the world’s highest level. The examples from recent weeks and months listed at the beginning illustrate what the consequences could be—and likely will be. And these examples concerned only criminal outliers from AI systems that were actually considered harmless. Outliers that mostly occurred without malicious intent and without provocation. Imagine what AI would be capable of if neither morality nor ethics played a role in its use, but only the result mattered. For example, in countries or alliances of countries that may not have the world’s best interests at heart.

However, it also remains a fact that artificial intelligence can be of enormous help to humanity. It accelerates learning and development processes. It simplifies repetitive tasks. It makes the economy, industry, and research more efficient. Like everything humanity has developed so far, it has its good and bad sides. It can enrich the world or become a threat to it. We can only hope that humanity will promote the positive aspects of AI in the future and work together to curb or prevent the negative ones.

Otherwise, Goethe and *The Sorcerer’s Apprentice* will likely come into play once again.

– – – – –