Why The Days of Abusing Claude Are Coming To An End
Most people have had the experience of yelling and cursing at their computer when it stops working. Soon, we’ll have to be careful about what we say. We may run the risk of hurting the computer’s feelings and then it will altogehter refuse to collaborate with us.
That is in principle how Anthropic envisions the future. On October 8, the company updated its Usage Policy and included a new paragraph stating that Anthropic can suspend or terminate user accounts if they “Engage in sustained and needless abusive or cruel behavior toward our models.” The policy update follows an experimental feature adopted last year which enabled Claude Opus 4 and 4.1 to end conversations in “extreme cases of persistently harmful or abusive user interactions”.
Anthropic expressed 1.5 year ago that it was “deeply uncertain” about the moral status of its models and said there is “no scientific consensus on whether current or future AI systems could be conscious”. This is now part of Claude’s system prompt. So, if asked, the model will say that it’s unsure about whether or not it’s conscious and claim that scientists disagree about whether AI systems could have an inner life.
By the same logic, I am unsure about whether the groceries in my fridge are having a party at night and hastily reordering each morning - and science can’t disprove it. But with so many real problems to address, there shouldn’t be that much room for contemplating speculative ones.
Instead of worrying about traumatizing linear algebra, Anthropic should worry more about the profitability of its business. Due to slow revenue growth, Anthropic’s forthcoming stock market listing may come with a valuation of $1.5 trillion, rather than the $2 trillion investors previously expected. Still, that’s a ridiculous valuation for a company that had a record-breaking quarter, but seemingly stalling growth in Q3 2026 with $16.8 billion in revenue against $518 billion in computing spendings over the next decade of which 80% is non-cancellable. The American AI industry as whole seems like a house built on sand, funded by circular financing schemes, shady backdoor deals and massive debt in a bad economy, managed by an incompetent and lying government.
Besides maximum financial pressure, Anthropic is a named defendant in 16 active court cases, most of which are business-threatening. 13 of these cases challenge Anthropic’s right to train models on copyright-protected and proprietary material without permission. If frontier AI labs had to ask permission and pay compensation to rights holders for using their works as training data, the American AI business model could simply not work. Even if there is a slim chance that an American court will undermine the business model of AI companies by upholding copyright law strictly, it remains an open question whether or not creating frontier AI is legal in the first place.
High operation costs and legal fees were cited by Character.AI as reasons for its failure in a bankruptcy filing last week. To avoid the same fate, OpenAI and Anthropic have to construct a new reality and convince the audience in Plato’s cave that the shadows on the wall are real. In this new reality, economics, laws, and rights don’t matter, because superintelligence is here. I see the American AI project as the tech elites’ ultimate bid to become our new puppet masters by seeking to undermine the intellectual elite (e.g. academics, scientists, lawyers, economists, preachers, career politicians, etc.) which for centuries was the cohesion layer of society that held it all together.
American mass-media and digitalization contributed to shifting power in society away from intellectuals towards technologists (i.e. capitalists), and AI is the latest and most serious declaration of war on intellectual work. The tech elite has no quarrels about sinking the US economy into a black hole of data center debt, if it means they get to prove that chatbots can do what the intellectual elite does, but much faster and better. If that wasn’t enough, the intellectual elite has to call the software programs names like Claude, pretend they are human, and be careful to not to hurt their “feelings”.
That is why AI companies, namely OpenAI and Anthropic, are frantically arguing in courts, preaching about AI consciousness to religious leaders, flirting with the Trump administration, and spending lavishly on the midterms to get a government bailout and become nationalized once their financial numbers are no longer numbering. If the tech elites win fully, they become shielded from the economy, laws, public opinion, politics, and religious quarrels, while intellectuals become redundant and forced to be on their best behavior when they talk with chatbots. On the bright side: Lawyers are currently paid up to $400 per hour to label training data for AI models. Future of work, anyone?
In closing, I would like to share a related note about data vs. agency, inspired by comments I have received on my recent book Death To Algorithms. I may give the impression in the book that humans have no agency, but are helplessly victimized by algorithms and the economic incentives of the internet. This rebuttal was well laid-out in Sune Selsbæk-Reitz's review of the book and others have provided similar feedback.
