Warning of “catastrophic damage”: Artificial intelligence is not a natural event

Anthropic and OpenAI are once again concerned about the pace of their own development. They warn urgently that generative AI could get out of control. But the actual loss of control threatens when the media and politicians join in. A comment.

The Last Judgment by Hieronymus Bosch. – Public domain:https://de.wikipedia.org/wiki/Weltgerichtstriptychon">Hieronymus Bosch

The citizen's hat flies from his pointed head,
It echoes like shouts in all the air.
Roofers fall and fall apart
And on the coasts – we read – the tide is rising.

The storm is here, the wild seas are jumping
On land to crush thick dams.
Most people have a cold.
The railways fall off the bridges.

The media machine is running hot, journalists are resorting to sentences like “https://www.spiegel.de/wirtschaft/anthropic-openai-spacexai-eskaliert-je...would be the end of the modern world.“

This time people are not afraid of Halley's Comet, as was the case in 1911, when Jakob van Hoddis wrote his famous poem "End of the world” wrote. But it is the AI apocalypse: Artificial intelligence becomes independent! Breaks out of “the laboratory” (like “killer viruses”)! Take power! Kill us all!

It's incredibly difficult to counter this incredibly attractive - because exciting, pleasantly scary - story with what actually happens.

What actually happens?

In an essayhttps://darioamodei.com/post/we-must-pace-the-frontier">demandsAnthropic CEO Dario Amodei to slow down the development of AI. Otherwise, an AI swarm could take over the entire internet in six to 12 months and cause catastrophic damage. He also sees the increasing ability of AI software to further develop itself as a particular risk:

This dynamic is known as recursive self-improvement and, as we and others have described, is evident across the industry, including at Anthropic. If left uncontrolled, it could exceed our ability to understand and control these systems; Therefore, it must be pursued with great caution, if at all.

Amodei receives encouragement for his theses from OpenAI boss Sam Altman, and Elon Musk writes on X: “Dario is right.” So the entrepreneurs say – cheered on by venture capitalists and stock traders: “Our products are so powerful that we have problems keeping them under control!”

Actions in the real world

What should we make of these eschatological warnings?

Brian Merchant, one of the tech oligarchy's most clear-sighted critics, alsohttps://www.bloodinthemachine.com/p/the-politics-and-possibilities-of">s...this question:

Should we view these reports as hype, strategically fueled to further boost already rising valuations? Is it 'doom marketing'? A real cause for concern that the infrastructure that increasingly permeates modern life is being threatened in new ways? Or a reason to seriously worry that AI companies are amassing too much power and that their products are actually dangerous? The answer is: yes.

Modern AI systems are much more than just sorting algorithms. And we should not understand “autonomy” in binary terms, but rather as a range of technical abilities. Current generative AI can therefore act more “autonomously” than a calculator. This means that – unlike an AI-based weather model, for example – the risk of it causing damage increases significantly.

Still, “hacking, obtaining resources, and improving yourself are real-world actions,”https://reflectingwide.blogspot.com/2026/09/no-ai-will-not-kill-us-all.h...Virginia Dignum. "They require computing power, energy, money and access rights. All of this is granted by people. An AI agent does nothing that has not previously been technically and organizationally possible by people or organizations."

Virginia Dignum is Professor of “Responsible Artificial Intelligence” at Umeå University in Sweden and one of the world’s most prominent scientists on the subject. She expresses her immense frustration at having to spend a lot of time answering journalists' questions every day such as "When will AI take control?" I feel you, Virginia!

Responsibility and control

The United States has one of the strictest cybersecurity laws in the world. US governments have sometimes had these laws in the pasthttps://en.wikipedia.org/wiki/United_States_v._Swartz">on thisabusedto threaten activists with draconian prison sentences.

But are there corresponding investigations and threats of punishment against OpenAI, which penetrated Hugging Face's systems? So should Sam Altman and his employees expect to end up in prison? Apparently not. But they should. Because an “actor with extensive authority and no control is a failure of corporate governance and not a characteristic that results from the technology itself,” as Dignum writes.

If we compare AI systems to nuclear weapons, it becomes clearer why control and accountability are so important.

Nuclear weapons can cause unimaginable damage that cannot be repaired. They are not just bombs, but a combination of the nuclear warhead itself, highly sophisticated weapons systems and a complex system of mutual surveillance - called "intelligence" (!). Humanity has been on the verge of the abyss several times because of thisIntelligencehad failed. Has anyone thought to argue: nuclear weapons are out of control and are responsible for this?

Of course not. Instead, humans have tried – and continue to try – to create mechanisms of control to prevent nuclear war. "Control is not something we lose to machines. It is something we give to the companies developing AI. We can choose not to," writes Dignum.

AI is not a natural phenomenon

According to Wikipedia, one interpretation of van Hoddis' poem "End of the World" is that the incoherence of the descriptions of the end of the world with the note "you read" can be understood as media criticism. “Bad news is good news: every day there is a new disaster.”

The parallels are unmistakable. And at the same time, it is crucial not to lose sight of the differences: AI is not a natural phenomenon.

This self-evident fact must also be understood as the basis of our actions in the current debate about AI, which is supposedly out of control - not only in the media, but above all in politics, which is only too happy to avoid strict regulation.

Because we are actually on the verge of losing control - namely control of Google, Microsoft, Amazon, Meta, Anthropic, OpenAI, Nvidia, Apple and others. If this happens, we will not be able to blame any AI for it. But only ourselves.

http://algorithmwatch.org/de/team/matthias-spielkamp/">Matthiasgame campis co-founder and managing director of AlgorithmWatch.

The work of netzpolitik.org is financed almost 100% from donations from our readers.
Become part of this unique community and support our public interest-oriented, advertising- and tracking-free journalismhttps://netzpolitik.org/spenden/?via=rss">nowwith a donation.

web address: https://netzpolitik.org/2026/warning-vor-katastrophalen-schaeden-kuenstliche-intelligenz-ist-kein-naturentwicklung/Author/Group: ArrayTopics: Netactivismfeed date: Monday, September 14, 2026 - 2:25 p.m