2026
Anthropic's dispute with the Defense Department tests whether its bans on mass domestic surveillance and autonomous weapons are enforceable principles or just sales friction.
2026
The Responsible Scaling Policy stopped reading like a unilateral stop rule and started reading like a flexible internal framework, which matters because the company sold itself on stronger self-restraint than most peers.
2025
Judge Alsup helped Anthropic on lawfully acquired books while preserving a much uglier record around the pirated central library, data provenance, and the gap between public responsibility branding and internal practice.
2025
Anthropic's growth appears real, but annualized consumption snapshots, channel opacity, and compute strain make the headline numbers less comparable and less durable than they first look.
2024
Anthropic established real model strength, but the investor question was never whether the numbers existed. It was whether tool scaffolding, benchmark choice, and later footnotes made the story cleaner than the operating reality.
2023
Anthropic built a governance structure more thoughtful than startup boilerplate and much softer than believers pretend. The legal architecture is meaningful, but far less constraining than the brand implies.
2021
The founding claim about reliable, steerable systems was not fake. The five-year record is a study in what happened after that promise met hyperscaler capital, enterprise demand, and defense work.
When a corporation elects to drape itself in the vestments of moral authority, to raise billions not on the promise of profit alone but on the claim that it, uniquely among its competitors, can be trusted with the most consequential technology of the century, it does not merely invite scrutiny. It demands prosecution.
Anthropic's founders did not slip quietly out of OpenAI over stock options or office politics. They departed, loudly and publicly, on the declared grounds that the race toward artificial general intelligence required a steadier, more principled hand at the wheel. Their hand, as it happened. What follows is a specific and evidenced examination of the distance between that founding promise and what came after: the governance structures announced as bulwarks and functioning as furniture, the safety commitments drafted with great fanfare and quietly fed into the shredder when the contracts arrived.
The question is not whether Anthropic is worse than its rivals. It may not be. The question is whether a company that sold the world its conscience ever actually had the inventory.
Read Merchants of SafetyAnthropic may be a real company with real revenue and real technical accomplishments. It may also be a terrible investment. These two facts coexist more comfortably than the prospectus would suggest. The public story is cleaner, nobler, and more tightly managed than the business underneath it, and the distance between the two is where ordinary investors lose their money.
Our five-year analysis tracks what actually happened from 2021 to 2026. The cautious disclosures gave way to narrative management. The safety commitments aged like milk. The release behavior drifted, quietly and then not quietly at all, from every prior public commitment. We mapped the divergence so you would not have to take anyone's word for it, including ours.
Read the full analysis