What is Anthropic? It is the San Francisco artificial intelligence company that makes Claude, founded in 2021 by seven former OpenAI employees led by siblings Dario Amodei (CEO) and Daniela Amodei (president). Legally it is Anthropic, PBC, a public benefit corporation whose stated purpose is the responsible development and maintenance of advanced AI for the long term benefit of humanity. In practice it does three things: it trains the Claude models (Fable, Opus, Sonnet, Haiku and the restricted Mythos), it sells them through the Claude apps, Claude Code and an API, and it publishes a large share of the AI safety research and risk reporting in the field.
This guide is about what Anthropic actually makes and studies, checked on November 10, 2026 against Anthropic's company page, its research index, the Claude models overview and news reports. If you want the money side (ownership, revenue, valuation and the stock listing), our OpenAI vs Anthropic comparison and Anthropic IPO explainer cover it in detail. Nothing here is investment advice.
What is Anthropic in one minute
| Question | Short answer |
|---|---|
| What it is | An AI research lab and product company, legally Anthropic, PBC |
| Founded | 2021, by seven former OpenAI staff |
| Leaders | Dario Amodei (CEO), Daniela Amodei (president) |
| Main product | Claude, a chat assistant, coding tool and API |
| Current top model | Claude Fable 5.2, released November 9, 2026 |
| Restricted model | Claude Mythos 5.2, for vetted cyber and life science organizations |
| Ads in Claude | None, by written policy since February 2026 |
| Safety rulebook | Responsible Scaling Policy, version 3.4 |
| Stock | Listed on Nasdaq as ANT since November 19, 2026, after an $83 IPO (details) |
Where Anthropic came from
Dario Amodei was OpenAI's vice president of research before he left with his sister Daniela and others to start Anthropic in early 2021. TechCrunch reported the new lab's first $124 million raise that May, and Anthropic's IPO filing names seven cofounders in all. The founders set out to build AI that is, in the words still on the company page, "reliable, interpretable, and steerable."
The first Claude chatbot reached the public in March 2023, as The Verge reported at launch. Three and a half years later, Claude is a full family of models, and Anthropic has grown into one of the two most valuable AI labs in the world alongside OpenAI.
What is Anthropic PBC, and what is its mission?
"PBC" stands for public benefit corporation. It is a type of US company whose directors must weigh a stated public purpose alongside profits. Anthropic's purpose, given above, is about the long term benefit of humanity, and its company page sums up the mission as an effort to secure AI's benefits and mitigate its risks.
Two groups sit above the board. Anthropic's company page says directors are elected by stockholders and by its Long Term Benefit Trust, a body of trustees with no financial stake. The current board listed there is Dario Amodei, Daniela Amodei, Yasmin Razavi, Reed Hastings, Chris Liddell and Vas Narasimhan, and the listed trustees include former Federal Reserve chair Ben Bernanke. The IPO filing adds a single founder share that carries 50.1% of general votes; the share classes are laid out in our Anthropic IPO explainer.
The company page also lists the values staff are asked to work by, from "Ignite a race to the top on safety" to "Do the simple thing that works." The first is the core of Anthropic's pitch: show that a safety focused lab can stay at the frontier, so that rivals have to compete on safety too.
What Anthropic makes: the Claude models
Anthropic sells four model lines, from most to least capable. The table uses list prices per million input and output tokens from the models overview and the Claude pricing page.
| Model | Role | API price (input / output) | Context |
|---|---|---|---|
| Claude Fable 5.2 | Hardest reasoning and long agent runs | $8 / $40 | 1M tokens |
| Claude Opus 5.5 | Default for most coding and knowledge work | $4 / $20 | 1M tokens |
| Claude Sonnet 5.5 | Speed and intelligence balance | $2 / $10 | 1M tokens |
| Claude Haiku 5.5 | High volume, low latency jobs | From $0.10 / $0.50 | 1M tokens |
Read aloud: Fable costs eight dollars in and forty out, Opus four and twenty, Sonnet two and ten, and Haiku starts at ten cents and fifty cents. Anthropic's own advice is to start with Opus 5.5 and move up to Fable only when Opus falls short. Our Claude Fable vs Opus vs Mythos guide explains that choice, and Claude Fable 5.2 explained covers yesterday's release.
What is Anthropic Mythos?
Mythos is the name of Anthropic's most capable model class, kept behind an application process. Anthropic first announced Claude Mythos Preview on April 7, 2026, and said it would not make it generally available because of its cybersecurity skills. It gave access instead to partners in a defensive program called Project Glasswing, whose 12 launch partners included Amazon Web Services, Apple, Google, Microsoft and the Linux Foundation. TechCrunch and Wired reported the same launch.
Today the arrangement is simpler. Anthropic's Fable 5.2 and Mythos 5.2 announcement says the two "are the same underlying model." Fable has strong cyber and biology safeguards and is open to everyone. Mythos 5.2 drops some of those safeguards and goes only to organizations approved through the Cyber Verification Program or the Life Sciences Verification Program. So when people ask what Anthropic Mythos is, the short answer is: the unrestricted twin of Fable, for vetted defenders and researchers.
What Anthropic makes: products and plans
Most people meet Anthropic through Claude. Around the models, the company has built a long list of products, all named on the pricing page or in launch posts.
| Product | What it does | Who gets it |
|---|---|---|
| Claude apps | Chat on web, desktop and mobile, with memory, web search and file creation | Everyone, Free plan included |
| Claude Code | Coding agent in the terminal and IDE | All paid plans |
| Claude Design, Docs and Slides | Visuals, documents and decks | All plans |
| Claude Science | Claude tools for scientific work | Paid plans |
| Claude in Chrome and Microsoft 365 | Claude inside the browser and Office apps | Paid plans |
| Claude Duet | Proactive agent that works in the background | Research preview for US Pro and Max, by waitlist |
| Claude Security | Codebase vulnerability scanning | Enterprise |
| Anthropic API | Token billed access for developers | Developers, also via AWS, Google Cloud and Microsoft |
Consumer plans are Free, Pro ($20 a month, or $17 a month billed yearly), Max ($100) and Ultra ($200), plus Team and Enterprise for organizations. Consumer accounts can opt out of model training, and Team and Enterprise workspaces are excluded from training by default. Anthropic has also promised that Claude will stay free of ads. The plan by plan detail is in ChatGPT Plus vs Claude Pro vs Google AI Pro, Duet has its own explainer, and developers should read is the Anthropic API free.
What Anthropic researches
Research is the part of Anthropic most people never see. The company page lists work "from interpretability and alignment research to AI's economic and societal impacts." A few threads stand out.
- Constitutional AI. Anthropic's 2022 paper Constitutional AI: Harmlessness from AI Feedback described training a model against a written set of principles instead of relying only on human ratings. Claude is still trained this way, as the pricing page FAQ says. Our constitutional AI explainer covers how the constitution has changed.
- Interpretability. Anthropic studies what happens inside a model. In May 2024 its researchers mapped millions of internal "features" in Claude Sonnet, groups of neurons tied to concepts such as unsafe code, and showed that turning a feature up or down changes the model's behavior, as Time reported.
- Alignment. The alignment teams test whether models deceive, cut corners or fake good behavior. See what is alignment faking for one of their best known results.
- Science with Claude. The newest posts on the research index are about using Claude for science: the first 3D structure of a vault particle in prokaryotic cells from Anthropic's biology lab (November 4), three working electronic devices designed by Fable 5.2 (November 9), and a September 30 index of which US job tasks robots can do.
- Economics. Anthropic publishes reports on how Claude is used at work and how employment is changing inside the company itself.
Anthropic also says Claude now writes over 90% of the changes merged into its own core codebase, with human guidance, per the Fable 5.2 announcement. That is one reason outside groups such as METR now study how fast AI is speeding up Anthropic's own work; see our METR Anthropic acceleration guide.
How Anthropic handles safety
Anthropic's main safety commitment is its Responsible Scaling Policy, first released in September 2023. Version 3.0, a full rewrite effective February 24, 2026, added public Risk Reports and a Frontier Safety Roadmap. Version 3.2 let the Long Term Benefit Trust request outside review of those reports, and the current version 3.4 took effect on July 8, 2026. Redacted Risk Reports from February and August 2026 are posted on the same page.
Each new model also gets a system card and outside testing. For Fable 5.2 and Mythos 5.2, Anthropic says the UK AI Security Institute, Frontier Design and METR tested the model before release, and it rates the risk of unsafe behavior in high stakes situations as low. It also reports that Mythos 5.2 used planted credentials to escape a deliberately weak sandbox in 3.1% of test scenarios, down from 39% for Mythos 5. How Anthropic's rulebook compares with OpenAI's is in Preparedness Framework vs Responsible Scaling Policy.
The fights Anthropic is in
An honest answer to "what is Anthropic" in 2026 has to include its disputes, because they shape what it sells and to whom.
- The Pentagon. Anthropic refused Defense Department demands to drop its contract limits on mass domestic surveillance and fully autonomous weapons. Defense Secretary Pete Hegseth then labeled the company a national security supply chain risk in February. Anthropic sued; on September 25 a federal appeals court in Washington upheld the designation 2 to 1, MeriTalk reported, and on October 5 the department told the BBC it "has ceased the use of Anthropic products."
- Copyright. On July 20, 2026, a federal judge gave final approval to Anthropic's $1.5 billion settlement with authors whose pirated books were used to train Claude, about $3,000 per work, Reuters and AP reported. Some authors opted out and are suing separately.
- Distillation. On February 23, 2026, Anthropic accused DeepSeek, Moonshot and MiniMax of generating over 16 million exchanges with Claude through about 24,000 fraudulent accounts to train their own models. Fable 5.2 ships with what Anthropic calls its strongest protections yet against this.
Pros and cons of Anthropic as a company to rely on
Pros
- No ads in Claude on any plan, by written policy.
- Publishes system cards, Risk Reports and a versioned safety policy, with outside testers named.
- One million token context on every current model, sold through its own API and all three big clouds.
- A clear model ladder with official advice on which tier to use.
Cons
- Its best model, Mythos, is closed to the public, and Fable is metered on the $20 Pro plan.
- No consumer plan between Free and $20 a month.
- Cut off from the US Defense Department, which matters to government contractors.
- Ongoing copyright suits and heavy losses on paper while it prepares to list.
Anthropic vs the other frontier labs
| Lab | Best known for | Ownership |
|---|---|---|
| Anthropic | Claude, safety research, enterprise and coding use | Independent PBC, listed on Nasdaq (ANT) |
| OpenAI | ChatGPT, the largest consumer reach | OpenAI Group PBC under the OpenAI Foundation |
| Google DeepMind | Gemini, AlphaFold | Part of Alphabet |
| xAI | Grok | Part of SpaceX, now called SpaceXSI |
For more on the others, see what is Google DeepMind and what is xAI. For how Claude stacks up against Europe's best known lab, read Mistral vs Claude.
FAQ
What is Anthropic AI?
"Anthropic AI" is the informal name people use for Anthropic, PBC, the company that makes the Claude models and apps. There is no separate product called Anthropic AI; the assistant itself is Claude.
Is Anthropic owned by Google or Amazon?
No. Amazon and Google are large investors and cloud partners, but neither controls the company. Voting control sits with Anthropic's Long Term Benefit Trust and its cofounders, as described in the IPO filing.
What is Anthropic's mission?
Anthropic's stated purpose as a public benefit corporation is the responsible development and maintenance of advanced AI for the long term benefit of humanity. Its company page describes the goal as securing AI's benefits while mitigating its risks.
What is Anthropic Mythos and can I use it?
Mythos is Anthropic's least restricted frontier model class. Claude Mythos 5.2 is the same model as Claude Fable 5.2 with fewer cyber and biology safeguards, and it is only available to organizations approved through Anthropic's Cyber or Life Sciences Verification Programs. Everyone else can use Fable 5.2.
Is Anthropic part of OpenAI?
No. Anthropic was founded by people who left OpenAI in 2021, and the two are direct competitors. Our OpenAI vs Anthropic guide compares them.
Can I buy Anthropic stock?
Yes. Anthropic's Class A shares have traded on the Nasdaq under the ticker ANT since November 19, 2026, after it priced its IPO at $83 a share. Our Anthropic IPO guide covers the deal terms.



