A safety-first AI firm that paid $1.5bn for pirated books—and then built a machine to scan every book in the world.
Anthropic positions itself as a safety-first AI builder—but its business runs on proprietary models trained with legally contested data, sold under restrictive partnerships, and governed by self-declared public benefit terms that do not prevent mass-scale book scanning or billion-dollar copyright liability.
Anthropic is a San Francisco–based public benefit corporation founded in 2021 by ex-OpenAI siblings to build safe AI.
2:50
What They Sell
It sells four tiers of proprietary LLMs via chatbot, API, and vertical services—with Mythos locked to US cybersecurity and life sciences partners.
4:45
Where the Data Comes From
Before Project Panama, Anthropic used pirated books—and paid $1.5 billion to settle the resulting 2025 copyright lawsuit.
6:24
Safety vs. Surveillance
Its stated mission is AI safety—but its largest contracts are with the DoD and Intelligence Community, under narrow restrictions.
Worth your time?
Yes. Study the whole thing.
3.5/ 5
What works
tiered model commercialisation
government contract segmentation
narrative discipline around safety
What does not
anthropic's public benefit corporation status constrains its data acquisition practices
Study it if
developers evaluating LLM vendors
policy teams assessing AI procurement risk
legal teams reviewing third-party model licensing
Skip it if
investors seeking revenue or margin disclosures
researchers expecting open weights or reproducible training logs
The written brief1 min read
What the company or idea is
Anthropic is an American AI public benefit corporation founded in 2021 in San Francisco by Dario and Daniela Amodei to promote AI safety.
How it actually makes money
Anthropic makes money by selling access to its Claude models via API, chatbot subscriptions, and domain-specific services like Claude Code and Cowork.
What works
Its tiered model strategy (Haiku, Sonnet, Opus, Fable) and restricted high-power deployment (Claude Mythos for US cybersecurity and life sciences partners) create defensible commercial segmentation.
What does not
Its public benefit corporation status does not constrain its data acquisition practices: it settled a $1.5 billion copyright lawsuit in 2025 after using pirated books, then launched Project Panama to destructively scan all the world’s books.
What to take from it
The gap between Anthropic’s safety-first narrative and its litigation-driven data sourcing reveals how regulatory and reputational risk management—not technical alignment—shapes its operational boundaries.
Is it worth your time
Yes—if you need a commercially viable LLM with government-contracted guardrails, but no—if you expect transparency on training data provenance or alignment claims.
