Skip to content
Welt am Draht
DE
The new fascismArticle draft

Power without oversight: why the concentration of power at AI companies is a question of democracy

Some of the men who build and fund AI consider democracy dispensable – and they have allies in governments. We call this a fascist process, and we disclose the criteria we use.

By Georg Else R. · · 16 min read

LinkedInE-mail
Somber steel-blue scene: a wire globe hovering over a checkerboard floor, next to a monitor showing a face and a camera on a tripod.
A wire globe above the checkerboard floor: power that observes without being seen itself.Image: Original AI-generated concept image made for this prototype

Core claim & position

Core claim
A few corporations control data, computing power and platforms. Where this power joins forces with authoritarian politics, it is a democratic problem – regardless of the label.
Position
We commit to it and call the documented development a fascist process – as a concept of process, not a description of a state of affairs. The evidence is in the text, and we say where it ends.

“Künstliche Intelligenz und der neue Faschismus” (“Artificial Intelligence and the New Fascism”) is the title of Rainer Mühlhoff’s book, published by Reclam in July 2025 and awarded the Friedrich Ebert Foundation’s “Das politische Buch” prize in 2026. His thesis: AI is an instrument of power in the hands of a few corporations and is easily harnessed for authoritarian purposes. This article takes the thesis as the starting point for a warning. It examines what can be substantiated and explains why we use the word “fascism” despite the objections.

01What Mühlhoff claims

Mühlhoff, Professor of Ethics and Critical Theories of Artificial Intelligence at Osnabrück University, does not regard AI as a neutral tool. In his view it is an instrument of power in the hands of a few corporations that command data, media and computing power. Without state regulation, he argues, this lends itself to authoritarian ends.

He identifies fascist traits in three features: antidemocratic action, a readiness for violence (including symbolic and verbal violence), and the deliberate harnessing of the latest technology as an instrument of power. According to the book digest on the first edition, he does not claim an automatism but speaks of fascistoid tendencies, which only the further course of events would turn into a system.

In the WDR 5 programme “Das philosophische Radio” (1 September 2025, in conversation with Jürgen Wiebicke), Mühlhoff explains how he arrived at the term and what he means by it. The idea for the book arose in November 2024, when he read Edwin Black’s book on IBM and the Holocaust. For him, fascism is above all a political attitude, not necessarily a political system or a genocide. Of the actors he talks about, nothing has been heard about planned extermination of the Jews, he says, and fairness requires adding that. In his assessment, genocide can still be averted. But among some political forces in the United States and in Europe he sees a great deal of this attitude: destructiveness, violence as a political means, the attack on the rule of law, and an almost erotic relationship to modern technology. He names Umberto Eco as one of his main sources; his three criteria are a slimmed-down version of Eco’s list.[1]

02What can be substantiated

First: influential figures openly question democracy. Peter Thiel, co-founder of PayPal and Palantir, wrote in the journal Cato Unbound in April 2009: “Most importantly, I no longer believe that freedom and democracy are compatible.” In context, he meant that majority decisions endanger freedom as he understands it, namely that of free markets. In October 2023, Marc Andreessen named “Trust and Safety” and “Tech Ethics,” among other things, as enemies in his “Techno-Optimist Manifesto.” VentureBeat sums up the text’s worldview as treating regulation as bad by default. The “Dark Enlightenment” around Curtis Yarvin and Nick Land explicitly wants to replace democracy with CEO-style rule. These are influential funders and idea-providers of the industry, not fringe figures.

Second: the industry is divided, and that is part of the truth. In May 2023, Sam Altman called for a national oversight agency for AI before the US Senate. According to Built In, Anthropic has publicly donated 20 million dollars to the side favoring stricter rules. On the other side stands “Leading the Future,” launched in August 2025 by, among others, Andreessen Horowitz and OpenAI president Greg Brockman, with millions from OpenAI. The group wants to support AI-friendly candidates and attack proponents of stricter rules. The claim that “nearly all AI chiefs are against regulation and democracy” therefore does not hold. What can be substantiated is this: an influential group with open distance from democracy and a great deal of money in the fight against concrete rules.

Third: the link to government power is documented. On 11 December 2025, Donald Trump signed the executive order “Ensuring a National Policy Framework for Artificial Intelligence.” The Department of Justice is to set up an “AI Litigation Task Force” that will go after the AI laws of individual states. States with laws the administration deems “onerous” are to lose funds from the BEAD broadband programme. Whether you call it industrial policy or a shift of power: rules that legislatures pass for AI are meant to be cleared out of the way.

Fourth: what this looks like in everyday practice. In March 2025, a software developer who had just started at DOGE writes a script. It is meant to sort contracts of the US Department of Veterans Affairs by which ones can be cut. An AI model reads only the first roughly 2,500 words of each contract. For more than 1,000 contracts it reads out a value of 34 million dollars, some of which were actually worth 35,000 dollars. The tool covers over 2,000 contracts and flags the dispensable ones as “munchable.” Among the flagged contracts that had already been cancelled at the time of ProPublica’s research in June 2025 is the maintenance of a genome-sequencing device used for cancer research. The Veterans Affairs department says contracts are cancelled only after several reviews by its own staff. By its account, then, a faulty tool supplied suggestions and humans decided. The case shows no plan, but sloppiness with authority, and that is dangerous enough.

Fifth: Mühlhoff’s further arguments from the conversation. He links corporate power to a logic of growth: if the industry keeps growing as it has, it will hit limits in electricity, water and data that a democratic politics could no longer sustain. Then democracy and the rule of law would have to be “pushed aside.” That is a forecast, not a fact; he himself calls estimates of electricity demand through 2030 uncertain. He also points to Palantir contracts with the Pentagon worth roughly 10 billion dollars and to the use of Palantir systems by the US immigration agency ICE. These claims come from the conversation and have not been verified by us.

Sixth: what Mühlhoff says about the idea-providers. In the WDR programme he calls Peter Thiel a convinced eugenicist who wants to abolish public universities and tailor education to industrial purposes. Eugenics is returning in tech ideologies, he says, for instance in the “IQ fetish”: the concept of IQ is itself a product of the eugenic pseudoscience of the twentieth century. He calls Effective Altruism the “gateway drug” to more radical tech ideologies, and he says money from Silicon Valley is flowing into student groups at German universities. On the Dark Enlightenment, he describes the idea of CEO monarchy: citizens are customers, they have no political voice, only the right of exit. These are his statements; we have not verified them individually. With Effective Altruism in particular, lumping it together is contested: critics of the “TESCREAL” concept accuse it of throwing different things together.

03Methods and milieus

Mühlhoff’s criterion of “antidemocratic action” can be pinned down through methods. The following points are substantiated to varying degrees; we mark this in each case.

Alt-right and swarm attacks. The Gamergate campaign, starting in August 2014, began as harassment of a game developer and grew into a wave of threats, publication of private data (doxxing) and fake calls to the police (swatting) against women in the gaming scene. According to Britannica, it served as a recruitment channel for the alt-right, Breitbart under Steve Bannon amplified it, and it paved the way for Pizzagate and later QAnon. How far a straight line can be drawn from this to today’s politics is a matter of interpretation.

Sabotage instead of debate. In the conversation on Jung & Naiv, Mühlhoff describes how the AfD uses parliamentary inquiries (“Kleine Anfragen”) to disrupt authorities and procedures and to “troll” them. He calls this antidemocratic. His dividing line does not run between left and right, but between actors who take part in a culture of debate and those who call that culture and the structures of the rule of law themselves into question. That is his assessment. The Administrative Court of Cologne did not, for the time being, see an overall picture of this kind characterizing the AfD.

Flood instead of argument. The phrase “flood the zone with shit” was said by Steve Bannon, Trump’s former chief strategist, to the journalist Michael Lewis in 2018. It refers to the strategy of delivering so much material so fast that the media and opponents cannot keep up. At the start of the second term, Stephen Miller deployed it, according to Wikipedia: 55 executive orders in 20 days. Mühlhoff also warns of mass political disinformation on social networks and in AI information spaces that influences elections. He calls this a systemic risk and points to the Digital Services Act. He names no evidence for individual coordinated campaigns in the conversation; we therefore do not claim any.

Salvation and end times. In the conversation, the longtermists’ narrative of the future is described as a secular eschatology: a promise of salvation with end times and rebirth that serves religious needs. We would use the word “cult-like” only once we can substantiate it for a specific group. What can be substantiated is the religious structure of the narrative.

Christian nationalism. In a piece from October 2024, the historian Kristin Kobes Du Mez describes a growing alliance of Christian nationalists, tech billionaires and oil money and calls Project 2025 a blueprint. She names the Texas billionaires Tim Dunn and Farris Wilks as funders. That is her analysis and not a consensus, but it is a lead worth following.

CEO monarchy. Curtis Yarvin told the New York Times in January 2025: “I don’t believe in voting at all.” He wants to run the government like a company, with a CEO at the top. In 2012 he proposed “RAGE,” “Retire All Government Employees.” According to CNN, JD Vance cited him on a podcast in 2021, Peter Thiel backed his company Urbit, and Marc Andreessen calls him a “good friend.” All three say they do not share all of his theses. Yarvin himself claims to influence mainly younger government staff; that is his own statement and unsubstantiated.

These points show milieus and methods. They are not evidence of a coordinated overall project, and that difference matters.

04Why we use the word

Mühlhoff defines fascism through three features and then finds them in Musk, Trump and the AI industry. FAZ reviewer Uwe Ebbinghaus calls this circular reasoning, according to Perlentaucher. Mühlhoff’s list also omits what is often central in the scholarly debate: an ideology of exclusion and the myth of national renewal. Sven Reichardt considers populism and authoritarianism too weak as concepts precisely because they leave out ideology and racism, and proposes “post-fascism.” Robert Paxton considers the term fitting after 6 January 2021, Jason Stanley sees no better word, and Roger Griffin calls the label a distraction.

With the broader criteria, the word holds for parts of the political right. Racist statements and policies by Donald Trump are widely documented, for instance in the overview of his racial views; as early as 1973 the US Department of Justice sued him for discriminating against Black housing applicants, and in 1975 it ended in a settlement without admission of guilt. For the AfD the legal situation is more precise, and it remains open: the Federal Office for the Protection of the Constitution had classified the party in 2025 as confirmed right-wing extremist. The Administrative Court of Cologne temporarily barred this on 26 February 2026 in summary proceedings. It saw sufficient certainty of anti-constitutional efforts within the party and found violations of human dignity in demands such as banning minarets and headscarves, but no characterizing overall picture. According to ZDF, the main proceedings are likely to take years. What is substantiated, then, is an element in the party’s programme that violates human dignity. The classification of the entire party is currently not, and anyone who cites it should state the court status.

For the AI corporations, the word fits only indirectly. We have not substantiated racist statements by corporate chiefs. Mühlhoff sees biologistic notions and thinking in terms of superiority in the scene, but in the conversation he bases this on the history of the concept of intelligence rather than on quotations. Contempt for democratic procedures is substantiated for one group. The bridge is the alliance: Musk led DOGE and wrote in a guest article in “Die Welt” at the end of December 2024 that the AfD was “the last spark of hope for this country,” and Thiel supported JD Vance in his campaign. That is a connection of capital, technology and authoritarian politics. We consider the name from the 1930s appropriate if it is understood as a concept of process and not as a description of a state of affairs.

Mühlhoff has commented on the term himself. In the conversation on Jung & Naiv (episode 818, 2 April 2026) he said, in paraphrase, that “fascism” in his book describes not an existing state but a process of radicalization: when AI corporations and right-wing ideology work together, society sails in that direction without fascism already ruling. That defuses the charge that the title claims too much. It answers the circularity objection only halfway: a concept of process needs criteria by which one can tell how far the process has advanced.[2]

We opt for the word, for three reasons. First, early warning: anyone who waits until everything looks like 1933 will recognize it too late. Second, the criteria: antidemocratic action and an ideology of exclusion are documented among parts of the political right, the harnessing of the latest technology for purposes of power among parts of the tech industry, and an alliance connects the two. Third, Mühlhoff’s concept of process, which does not claim a finished fascism.

We take the objection seriously. This is the smoke detector problem: one that goes off at every piece of toast eventually gets disconnected, and Griffin sees exactly this danger. So for us the rule is: we use the word for the process, not for every individual person and party, and each time we name the evidence.

We use the word for the process, not for every individual person and party – and each time we name the evidence.

— Editorial guideline of this essay

05What remains if you leave the word out

Then questions remain that need no history seminar. Three test questions for every AI system that co-decides about people:

  • Who decides in the end, and who can object?
  • Who sees the data, and who sees only the result?
  • Who bears the harm when the system is wrong?

Regulation is contested here. In May 2026 the EU agreed to postpone obligations for high-risk AI to December 2027 and August 2028. Parliament approved in June; the formal approval of the Council was still pending at the time, and the later status has not been verified. Supporters see pragmatism in this, critics a watering-down.

06How the book was received

In 2026 the Friedrich Ebert Foundation awarded the book the prize “Das politische Buch” (10,000 euros; according to the foundation, an independent jury decides). Reviews praise its compactness and its overview of tech ideologies. The criticism comes from several directions: the FAZ considers the concept of fascism circular reasoning. Konrad Lotter in “Widerspruch” misses the question of ownership, and the Frankfurter Nachrichten find the empirical basis thin and the “What is to be done?” too general. Klaus Janowitz counters that AI can also be used emancipatorily. We did not find, in the reviews we examined, that the description of the concentration of power itself is disputed.

In the conversation on Jung & Naiv, too, the final chapter “What is to be done?” was described as the weakest part of the book. Mühlhoff pointed to the revised and expanded edition, in which this chapter becomes considerably more detailed, and said the focus of the first edition had been on naming the danger in the first place.

07Conclusion

The warning would stand even without a label; we attach one anyway. A few corporations determine how AI is built, some of their leaders have publicly questioned democracy, and a government is clearing rules out of their way. That is substantiated. We call this development a fascist process, in Mühlhoff’s sense and according to the criteria named above. Experts argue about it, and in the case of the AfD a court case is also pending. We consider it more honest to use the word and disclose the criteria than to avoid it.

Back to the Veterans Affairs script. Whether this case is called fascism is for experts to settle. The error itself needs no label. It needs someone who reads the whole contract.

08Sources

On the reception:

Notes

  1. [1]This paragraph relies on an automatic transcription of the broadcast, so we render Mühlhoff’s words in paraphrase.
  2. [2]The conversation was transcribed by machine and analyzed in English translation; statements are therefore rendered in paraphrase.

Further reading