Names, versions and owners change fast. Checked October 2026.
In plain words
Anthropic is a company, not a product. It was founded in 2021 in San Francisco by seven people who had worked at OpenAI, including the siblings Dario and Daniela Amodei.
Its main work is Claude: a family of large language models and the chat app built on them. Businesses also use Claude through Anthropic's API, through big cloud platforms, and inside other companies' products.
Anthropic is organised as a public benefit corporation, a company that is legally required to weigh a stated public purpose alongside profit. It describes its mission as helping the world safely make the transition through transformative AI, and it publishes research on AI safety and alignment.
Everyday example
A cooking school with a written rulebook
Imagine a cooking school that hands every trainee a short written rulebook: keep the kitchen clean, be honest about ingredients, never serve anything unsafe. Trainees cook, check their own dish against the rules, fix it, and learn from the fixed version. Anthropic trains Claude in a similar way, using a written set of principles it calls a constitution.
Where the example breaks: Trainee cooks understand why a rule matters. Claude learns patterns from examples of following the rules, so it can still misapply them or be talked around them in unusual cases.
Latestas of October 2026
- Chief executive
- Dario Amodei
- Current models
- Claude Haiku 5.5, Sonnet 5.5, Opus 5.5 and Fable 5.1
- Stock market
- Filed on 26 October 2026 for a proposed listing on the Nasdaq, under the symbol ANT
- Investors
- Include Amazon, which announced a further $5 billion in April 2026 on top of $8 billion earlier, and Google
These details change often. The rest of this page explains what it is, which stays true.
How it actually works
Anthropic is best known technically for a training method it published in December 2022.
Write down the principles
Anthropic writes a list of principles, its constitution, describing how the model should behave: be helpful, be honest, avoid causing harm.
The model critiques its own drafts
The model writes an answer, then is asked to critique that answer against the principles and rewrite it. The rewritten answers become training examples.
AI feedback alongside human feedback
In a second stage, the model compares pairs of answers against the principles, and those judgements are used for reinforcement learning. This complements RLHF, where people rate answers. Anthropic calls the method Constitutional AI.
Testing before release
Since September 2023 Anthropic has published a Responsible Scaling Policy, a framework for testing new models for dangerous abilities and adding safeguards as they grow more capable. It also runs red-teaming and studies how models work inside, related to explainability.
- Founded
- 2021, San Francisco, by seven former OpenAI staff
- Co-founders include
- Dario Amodei and Daniela Amodei
- Main product
- Claude models and the Claude app
- Structure
- Public benefit corporation; its board is elected by stockholders and a Long-Term Benefit Trust
Questions people ask
Is Anthropic the same as Claude?
Who founded Anthropic?
Seven former OpenAI employees founded it in 2021, including the siblings Dario and Daniela Amodei. Before Anthropic, Dario Amodei was a vice president of research at OpenAI. The box above shows who leads the company now.
Is Anthropic part of Google or Amazon?
No. Anthropic is an independent company. Amazon and Google are both investors and both supply it with cloud computing, but neither owns or controls it.
What is Constitutional AI?
It is Anthropic's method of training a model with a written list of principles. The model critiques and revises its own answers against those principles, and AI-generated judgements are used alongside human feedback to shape its behaviour.