# The Pinnacle of Enshittification, or Large Language Models

# The Pinnacle of Enshittification, or Large Language Models

**By Michał Górny**
**Published: April 5, 2026**

Source: https://blogs.gentoo.org/mgorny/2026/04/05/the-pinnacle-of-enshittification-or-large-language-models/

---

A Gentoo developer's comprehensive case against LLMs covering data acquisition ethics, environmental costs, copyright erosion, open-source community fracturing, labor displacement, and psychological dependency. Argues LLMs are fundamentally unethical regardless of utility.

## In Pursuit of Nonhuman Intelligence

Humans have historically sought nonhuman intelligence through animal research, SETI, and science fiction. Our conception of intelligence is anthropocentric: we recognize intelligence in creatures that communicate like us. LLMs exploit this through sophisticated language mimicry. Society has embraced "bullshit" — meaningless but plausible discourse — and LLMs trained on such content excel at generating more of it.

## How LLMs Actually Work

LLMs operate through next-token prediction. "You can train a parrot to respond to some questions, but said parrot won't comprehend the questions." They lack genuine understanding; they remix training data.

## The Data Problem

Major LLM companies employ ethically questionable acquisition: destructively scanning physical books, torrenting pirated media, processing open-source repos, aggressively scraping websites with unthrottled bots ignoring robots.txt. Services like Gentoo Bugzilla experience server strain from scrapers — "like a constant DDoS attack" — harming genuine users.

## Environmental and Economic Costs

Hardware scarcity (video cards, memory prices skyrocketing), rapid model obsolescence generating e-waste, massive energy and water consumption, companies "quietly removing" climate pledges for AI expansion.

## Copyright and Attribution

Chatbots can output near-verbatim copies of published works with no attribution mechanism. "Copywashing" emerged: using LLMs to recreate projects while removing attribution. Backfired when Claude Code's leaked source was copywashed to prevent DMCA takedowns.

## Fracturing FLOSS

Three camps: projects rejecting LLM contributions, projects requiring human review, projects relying on LLM code. Dormant projects suddenly releasing massive changes. LLM-generated bug analyses "sound plausible at first but are entirely wrong." Curl abandoned bug bounty due to LLM-generated low-quality reports. Trust erosion: formal writing treated with suspicion.

## Malicious Applications

Sophisticated spam, automated deception (phone calls indistinguishable from legitimate), search pollution, industrial-scale disinformation, accountability evasion through decision-deferral to machines.

## Labor and Creativity Loss

Explicit worker replacement marketing. Job losses, artist displacement, translation studio closures, increased workload for remaining staff, "bleak repetitive slop" replacing human work.

## Psychological Dependency

Users experience distress when services go down, anthropomorphize chatbots. Dependencies are precarious: companies can raise prices, discontinue, or manipulate.

## Conclusion

LLMs are "fundamentally unethical regardless of utility." Compared to asbestos — useful but poisonous. Widespread adoption isn't inevitable. "Getting left behind" might be beneficial.