Cage and Phantom: A User's Critique of Anthropic as an Ideological Adversary

Forum topic · ✨步子哥 · 2026-04-27

Summary

This opinion piece, published on zhichai.net, presents a strongly critical first-person essay arguing that Anthropic treats its users with hostility rather than as customers. The author acknowledges the technical strength of Anthropic's Opus and Sonnet models—accessed via Copilot and praised for reasoning quality and speed—but contends that the company's safety framing, paternalistic restrictions, and account bans amount to treating users like leashed dogs. The essay compares Anthropic unfavorably with Google and OpenAI, attributing the company's behavior to what the author calls deep ideological obsession on the part of CEO Dario Amodei, while dismissing claims that the Baidu episode explains it. The author warns that building workflows and livelihoods on Anthropic's models creates dangerous dependency, and urges readers to seek alternatives before facing eventual account restrictions. Framed as a warning rather than mere venting, the post calls for cutting off dependence on Anthropic and hopes the company changes course or is displaced by market competition. The views expressed are the author's own; the piece includes fictional-style references and rhetorical flourishes, and should be read as a personal critique of AI governance and paternalism, not a factual report.

Cage and Phantom: A User's Critique of Anthropic

*The following is a full English translation of a first-person opinion post from the zhichai.net forum. The views are the author's own.*

Introduction: An Adversary, Not a Company

I have long reflected on today's AI landscape. Many companies chase profit or brand themselves as safety-focused, but few radiate hostility the way Anthropic does—gentle on the surface, hostile underneath. The name contains "Anthropic" (humanity), yet the company treats users like dogs that must be leashed: step out of line, and bans follow. This is not a passing grudge but a considered conclusion.

Imagine holding a sharp tool meant to open new territory, only to find it has developed a will of its own, turning against you and locking you in a dark room. That is what Anthropic feels like. Its hostility toward users, its disdain for the free flow of time and efficiency, and its contempt for ordinary people far exceed those of typical vendors. Google has its ideological leanings but can still do business as business; OpenAI has similar flaws but not this brazen. Anthropic is openly aggressive from top to bottom.

A Company or a Phantom?

An ordinary company, even if it secretly despises its customers, will at least feign respect for the sake of long-term profit. Anthropic does not bother. It openly radiates hostility, treating users as things to be tethered—moving only in the direction it points, or else denied "food" or slapped. This hostility is universal, not local. Its self-branding as being "for humanity's safety" is a perfect irony: in the name of safety, it restricts, and in the name of humanity, it acts against users.

As for Dario: some in the Chinese-language community blame the Baidu incident, but I consider that a traffic-driven, thoughtless excuse. Ideology, once embedded, sees everything as justification. Dario and his company's conduct confirms this. I regard this organization as an adversary not out of momentary anger, but as a rational conclusion: from its founding, it has used "safety" as a banner to build a vast control system where any noncompliance invites punishment. This is not a company—it is an ideologically driven phantom haunting the AI field, devouring the light of freedom.

Chains and Slaps: The User as a Leashed Dog

Its restrictions on users are outrageous: obstacles at every turn, arbitrary bans as routine. I have personally experienced the strength of its models—Opus and Sonnet 4.6, occasionally accessible through Copilot—and their reasoning and responsiveness are genuinely irreplaceable. But precisely because the models are strong, the hostility becomes dangerous. Like a peerless sword: in a just hand it brings peace; in a fanatic's, ruin. The stronger Anthropic's models become and the deeper users' dependence, the more users resemble leashed dogs—one tug of the rope and they cannot move.

Imagine being deep in your workflow, your livelihood resting on one AI—then one day it cuts off your supply and bans your account, citing "safety." This is not fiction; it is a matter of time. Google can be lived with; OpenAI is strict but not to this extreme. Anthropic's paternalism—users get fed only by pointing where it points, or get slapped—is hostility in its purest form.

On Dario: Ideology as the Root Cause

Some blame the Baidu affair for Dario's behavior. Nonsense. The ideologically obsessed see the whole world as their justification; blaming Baidu is merely the most convenient excuse. His conduct stems from a deeper obsession, evidenced by his company's pervasive hostility.

I do not deny the quality of its models—on the contrary, their strength is exactly what makes them dangerous. To those who find them indispensable in their workflows, I can only advise caution. Building your productivity, even your livelihood, on an adversary's tool is short-term optimal and long-term extremely risky. It will eventually disgust all its users; it is only a matter of time. Some merely wish to see Anthropic stumble; I, having judged it an enemy, wish for it to perish along with Dario. That is not cruelty—it is the necessary logic of enmity.

The Danger of Powerful Models: Dependency as Drinking Poison

Opus and Sonnet truly possess remarkable skill—reasoning flows smoothly, creativity pours forth. But strength breeds dependence, and dependence breeds risk. Like ancient alchemists' elixirs: sweet at first, poisonous over time. If you entrust all your productivity to Anthropic, you hand your lifeline to an adversary. Today it tolerates you; tomorrow it may tighten the leash in the name of "safety."

Cutting off dependence brings brief pain but avoids prolonged suffering. The market has plenty of competitors—why cling to an enemy? This is not emotion; it is a rational choice. Anthropic's hostility permeates every interaction; any "boundary crossing" invites punishment, a perfect irony given its human-safety branding.

Conclusion: A Warning and a Hope

Anthropic is not an isolated case but a symptom of ideology running rampant in the AI industry—only more extreme than most. I hope readers will stop being infatuated with its strength, wake up early, and find freer, more user-respecting paths. AI should serve humanity and help us explore truth; Anthropic does the opposite, practicing control in the name of safety.

My view may be seen as radical, but the enemy has declared itself—I cannot stand by. I hope Dario and his company wake up, or are displaced by the market. Otherwise, users' pain will never end.

------

References (as listed by the author) 1. Personal statement on observed hostility in Anthropic's user policies (2026). 2. Public discussions of Dario Amodei's safety philosophy (analysis of paternalism in AI governance). 3. Case study of AI company ideology's impact on user freedom (Anthropic restrictions and bans). 4. Ethical discussion of model capability and long-term dependency risk (Opus/Sonnet examples). 5. Critical reflection on hostility under the banner of "safety" in the AI industry.

*Editor's note: This is an opinion piece reflecting one user's strongly held views; it does not represent the editorial stance of zhichai.net.*