ChatGPT and Claude Under Lock and Key: Why Europe Shouldn't Worry

When Donald Trump loves something, it's playing the strongman. Whether it’s about domestic, foreign, or economic policy, the US President is quick to resort to threats and measures like tariffs. On the other hand, the Trump administration wants to capitalize on its presumed lead in AI models like Claude, Gemini, or OpenAI's GPT series. To confront its arch-rival China, ChatGPT and similar models are to be spread across the globe. After all, the open-weights and open-source sector of the People's Republic is catching up significantly, thanks in part to state support.

The analysis platform Artificial Analysis currently ranks Z.ai's GLM-5.2 model in third place in the "Intelligence" category, ahead of Grok, Meta's Muse Spark, and the current Gemini Pro model. It’s no wonder that alarm bells are ringing in the White House. The latest development is a decree from early June that is supposed to grant the government preliminary access to large language models from US companies within a voluntary framework.

It’s not unrealistic to think that relevant authorities could prohibit a general release after such a preliminary review. After all, that’s what happened with Anthropic's Claude Fable 5. The mythos-class model was subjected to export controls by the US government just days after its launch, allegedly due to security concerns. So, it seems that Donald Trump doesn’t care as much about winning the AI race as one might think.

In the European tech sector, this development is causing nervous twitches. Especially since the White House now seems to be taking concrete steps. According to CNN, the US government has just asked OpenAI to roll out its next model only in phases. And without OpenAI and Anthropic, nothing happens in terms of AI and Europe. Or does it?

Access to Claude Fable 5 and GPT-5.6: Why No One Really Needs the Performance of Top Models

This fear of falling behind is indicative of the political influence of figures like Dario Amodei, Demis Hassabis, and Sam Altman. For instance, they sat as equal members among the heads of state at the G-7 summit, as if it were the most normal thing in the world. While Altman insists that tech companies should not be allowed to control AI, as reported by Deutschlandfunk, he would surely prefer it if ChatGPT were to permeate every facet of our lives.

Trump’s pushback is actually a blessing. It is indeed critical if states alone decide how technology is distributed. This further undermines Altman’s already absurd idea that AI is something that benefits all of humanity. But perhaps decision-makers will realize more quickly that neither ChatGPT, Claude, nor Gemini are the be-all and end-all.

These tools may perform well on benchmarks. Moreover, they may stir emotions and create alarm in business and politics through a mix of clever marketing and actual capabilities. But the fact is that these tools are completely overkill for most applications.

No company needs GPT-5.6 to organize its document storage. No corporation is necessarily reliant on the coding prowess of Claude Fable 5 to bring products to market. And the office workers who convert PDF files into PowerPoint presentations, thereby overstretching token budgets, as reported by 404 Media, certainly don’t need these powerful tools.

Dependency on OpenAI and Anthropic: Open Models Remain the Solution

What the AI companies from the US primarily sell is the narrative that it can’t be done without them. That the machine stands still without them. That there is an ever-increasing need for data centers and computing power to tackle the challenges of the future, environmental impacts notwithstanding, astronomical costs aside. It’s time to banish this idea from our minds.

Because not only Z.ai has caught up with GLM-5.2. Other smaller, more open models, whether from the US, China, or Europe, are becoming increasingly powerful and efficient. And open-source models are also improving. Of course, there are still applications in coding or research where OpenAI or Anthropic may have advantages, and the use of large models is relevant. But these advantages are shrinking.

And even in the field of cybersecurity, no one needs Mythos 5 to exploit vulnerabilities. Cybercriminals wouldn’t even have the budget for token maxing. And if less than half of the companies examined in a recent study by cyber insurer Baobab, with revenues of less than five million euros, use two-factor authentication, a local model running on a standard gaming GPU will suffice. Access to advanced AI models helps no one if the structures are lacking.

Even though it sounds like an old refrain to both experts and laypeople alike: We attribute the dominance of US frontier models to ourselves. Through complacency and a refusal to look left and right. My recommendation: Stay calm and perhaps consider reallocating token budgets for ChatGPT or Claude.

Depending on the use case and the size of the setup, switching to local solutions can pay off within just a few months, as researchers from Carnegie Mellon University have found. And those who use generative AI in the most sensible way—namely, in a measured and project-based manner rather than indiscriminately everywhere—might even benefit from it even faster.