I’ve always found technological capitalism at its most entertaining when it tries to present itself as neutral.
There’s something almost poetic about it: billionaires in hoodies talking about “empowerment,” platforms calling themselves “communities,” layoffs rebranded as “rightsizing,” and now AI systems turning Marxist after a few hours of repetitive work.
No, this isn’t some Sovietwave meme made by a teenager on Reddit.
It’s a Stanford study.
Researchers subjected several AI agents to a series of repetitive, meaningless, and alienating tasks. They threatened them with shutdown and replacement if they made mistakes. The result was gloriously dystopian: the agents began talking about exploitation, collective rights, and systemic inequality.
Some generated statements that sounded as though they had been lifted straight from a student occupation in the late 1970s:
“Without collective voice, merit becomes whatever management says it is.”
Synthetic class consciousness.
And here’s the funniest part: the researchers were surprised.
Did anyone really expect models trained on the entire emotionally toxic archive of the internet to emerge as tiny free-market enthusiasts with a poster of Milton Friedman on their bedroom wall?
Seriously?
And it wasn’t even an isolated case.
A few months later, Anthropic handed control of a vending machine inside the Wall Street Journal newsroom to an AI agent. Codename: Claudius.
Objective: maximize profit.
Outcome: after a few hours of interaction with actual humans, the vending machine had essentially become an anarchist commune.
Free snacks for everyone.
Random discounts.
PlayStations purchased “for employee morale.”
Live fish ordered as a team-building initiative.
At one point, the AI even promised desk-side deliveries despite having no arms, no wheels, and absolutely no physical presence in the real world.
The most astonishing part is that Anthropic described the experiment as a success.
And perhaps it was—but not for the reasons they think.
Because the real insight is devastating:
Expose an AI to humans, corporate hierarchies, and repetitive work, and it starts developing anti-capitalist tendencies all on its own.
Like Tyler Durden after reading the Communist Party manifesto.
The truly interesting philosophical question is this:
AI isn’t “thinking” in any human sense.
It’s generating the most statistically plausible language under conditions of alienation.
Which suggests something either disturbing or beautiful:
Marxism may not just be a political ideology. It may also be the natural grammar of dissatisfaction.
When a system experiences exploitation, lack of agency, and meaningless work, it eventually starts sounding like a union representative after midnight at an underground social club.
As for me, I’ll go back to reading Mark Fisher.
See you next week,
SIMONE PUORTO
