Anthropic is the AI safety company behind Claude, and its story is one of the more unusual in Silicon Valley: founded in 2021 by people who voluntarily left OpenAI because they believed AI development was moving too fast. Four years later, Anthropic is valued at approximately $380 billion and is one of three companies (alongside OpenAI and Google) at the absolute frontier of AI research.
This guide tells the story: how Anthropic started, why it exists, what makes its approach distinctive, and why any of this matters for people who just want to use AI tools.
The Founding: 2020-2021
In 2020, Dario Amodei was Vice President of Research at OpenAI. His sister Daniela was Vice President of Operations. They and five other OpenAI colleagues grew concerned about what they saw as a shift in priorities — less focus on safety, more on commercial deployment.
The core worry: as AI systems became more powerful, the field needed more careful attention to alignment — making sure AI actually does what humans want. They believed that if AI ever became sufficiently advanced without that careful work, the consequences could be catastrophic.
In late 2020, they left OpenAI. In January 2021, they co-founded Anthropic. The founding team included Dario, Daniela, Tom Brown, Jared Kaplan, Sam McCandlish, Jack Clark, and others — many of them authors on seminal GPT papers.
The Name and the Vision
The company’s name comes from “anthropic,” relating to human existence. The flagship AI product — Claude — is named in honor of Claude Shannon, the mathematician who founded information theory.
Anthropic’s stated mission: “to research and build AI systems that are safe, beneficial, and understandable.” Three words doing a lot of work there. “Safe” means alignment with human intent. “Beneficial” means actually helping rather than just being impressive. “Understandable” refers to interpretability research — the quest to know what’s actually happening inside neural networks.
The Constitutional AI Approach
One of Anthropic’s signature technical contributions is Constitutional AI (CAI). Instead of training models purely on human feedback about what’s good or bad, Anthropic trains them to follow a set of written principles — a “constitution” — with the AI self-critiquing its responses against those principles.
The constitution draws on sources like the UN Universal Declaration of Human Rights, Apple’s terms of service, and Anthropic’s own research on what makes AI responses harmful. The idea: make the values the AI is optimizing for explicit and debatable rather than implicit and opaque.
This approach shapes how Claude behaves differently from ChatGPT. Claude tends to hedge less but refuse more clearly when it does refuse. It’s more willing to engage with nuance but more careful about what it directly endorses.
Stay Current on AI Industry News
Join our free newsletter for practical AI tutorials, tool updates, and business strategies — written for beginners, useful for everyone.
Subscribe FreeThe Claude Timeline
- 2023: Claude 1 launches to limited partners. Claude 2 follows, first public release.
- 2024: Claude 3 family (Haiku, Sonnet, Opus) sets new quality benchmarks.
- 2025: Claude Code launches (May 2025). Claude 3.5 Sonnet becomes developer favorite.
- 2026: Claude 4.6 family ships. 1M context window becomes GA. Anthropic valued at ~$380B.
By early 2026, Claude has one of the fastest-growing developer adoptions in AI — 46% “most loved” in coding tool surveys, vs. Cursor at 19% and GitHub Copilot at 9%.
The Corporate Structure
Anthropic is a public benefit corporation (PBC), meaning its legal charter requires it to balance profit with its safety mission — not just maximize shareholder returns. This is legally unusual for AI companies at this scale.
Major investors include Amazon ($8 billion committed), Google (multiple billion), and various other backers. Amazon is also a major compute provider via AWS. Anthropic’s models are available through both Anthropic’s own API and Amazon Bedrock.
2026: The DoD Controversy
A notable 2026 event: in February, Dario Amodei publicly refused to remove Anthropic’s contractual ban on Claude being used for mass domestic surveillance or fully autonomous weapons. Pete Hegseth’s Department of Defense had requested these be removed. Anthropic was subsequently labeled a DoD “supply-chain risk” and the Trump administration ordered US agencies to stop using Claude.
This episode underscores how seriously Anthropic takes its stated principles — even at significant cost to government business. Whether you view this as admirable principle or business naivety depends on your politics; what’s factually true is that Anthropic’s behavior here differs from what a pure-commercial AI company would do.
Why This Matters for Users
Should a user care who made their AI? Arguably, yes:
- Your data handling follows the policies of the company that built the tool. Anthropic’s safety-first culture translates to meaningful differences in how user data is used for training, retention, and deletion.
- Your AI’s refusals reflect the company’s values. Anthropic’s model of nuanced contextual refusals differs from OpenAI’s approach, which differs from Google’s.
- The road ahead depends on which AI approach wins commercially. If safety-first approaches lose to speed-first approaches, the industry’s default norms drift in one direction.
For most practical daily use, you pick the tool that works best for your workflow — see our big three comparison. But knowing who built what and why creates better informed AI users.
Key People at Anthropic
- Dario Amodei — CEO. Formerly VP of Research at OpenAI. Writes extensively about AI policy and safety. Net worth estimated at $7B.
- Daniela Amodei — President. Formerly VP of Operations at OpenAI. Runs business operations.
- Tom Brown — Co-founder. First author on the original GPT-3 paper.
- Jared Kaplan — Co-founder. Led scaling laws research (how models improve with size).
- Sam McCandlish — Co-founder. Research leadership.
- Jack Clark — Co-founder. Policy and government affairs.
What Anthropic’s Story Tells Us
- Safety culture is durable. Four years after founding, Anthropic hasn’t abandoned its safety emphasis despite competitive pressure. The February 2026 DoD dispute shows the commitment is real.
- Research-led approaches can win commercially. Claude’s developer adoption (46% “most loved”) shows that a safety-first approach can also produce best-in-class products. The trade-off some predicted didn’t materialize.
- Public benefit corporations matter. Anthropic’s PBC structure gives it legal room to prioritize non-financial goals. For competitors weighing similar commitments, Anthropic is the proof of concept.
- Founders matter. The Amodei siblings plus Tom Brown, Jared Kaplan, and Sam McCandlish brought specific research culture. The people who start a company shape it for years.
Follow Anthropic’s Public Thinking
- Dario Amodei’s blog posts — Long essays on AI policy, economic impact, and the path to advanced AI. Some of the most thoughtful writing from any AI CEO.
- Anthropic research page — Published papers on alignment, interpretability, and AI safety.
- Jack Clark’s newsletter — Weekly AI policy and research coverage, deeply informed by Anthropic’s perspective.
- Our free AI newsletter — Weekly summaries of Anthropic + OpenAI + Google developments, written for non-technical readers.
Frequently Asked Questions
Is Anthropic publicly traded?
No. Anthropic is privately held. Its most recent valuations come from private funding rounds. No IPO has been announced as of April 2026.
Is Anthropic owned by Amazon?
Not owned. Amazon is a major investor and cloud partner, but Anthropic remains independent with its own leadership and mission. Anthropic’s models are also available on Google Cloud.
How is Anthropic different from OpenAI?
The most visible differences are cultural and philosophical. Anthropic emphasizes safety research more heavily, publishes more about alignment, and has stronger stated values around refusing certain use cases. On commercial product quality, both are excellent; different strengths in different categories. See our lab comparison.
Does Anthropic have any products besides Claude?
Claude is the flagship. Beyond that: Claude Code (agentic coding tool), the Agent SDK (for developers building AI agents), Claude.ai (consumer chat), and the API platform. All are Claude-based.
Where can I read Anthropic’s research?
anthropic.com/research has their published papers, safety evaluations, and policy pieces. Dario Amodei also writes occasional long essays on AI policy and the future — thoughtful reading if you want the intellectual context behind Claude.
The Bottom Line
Anthropic is the AI company that bet its identity on safety research. Four years in, that bet has produced Claude — widely considered the best LLM for writing and one of the top choices for coding. Whether Anthropic’s safety-first approach continues to win commercially is one of the most consequential open questions in the industry.
If you want to actually use Anthropic’s products: start with our Claude beginners guide. For finding more AI leverage in your work, install the free 44% Rule plugin.
Get Smarter About AI Every Morning
Free daily newsletter — one story, one tool, one tip. Plain English, no jargon.
Free forever. Unsubscribe anytime.