Anthropic is an American artificial intelligence (AI) company. It builds advanced AI models, turns them into products, and studies how to make AI safer. So what does Anthropic do in one line? It builds the technology, and Claude is its best-known product.
Many people first hear about Anthropic through Claude, an AI assistant that can write, explain, research, and help with code. But the company does more than run a chatbot. It sells AI tools to developers and businesses, studies how AI works on the inside, and works with U.S. government agencies. It also has an unusual legal structure meant to protect its public mission.
This guide explains what Anthropic does in plain English. You will learn about its products, its research, how it makes money, and how it handles your data. It also covers the main debates around the company, using details from its public filings and announcements.
What Is Anthropic?
Anthropic, PBC, is an AI research and technology company based in San Francisco, California. It was founded in 2021 and builds its own AI models. It then offers those models through its own apps, a developer platform, and business partnerships.
The company’s work falls into three main areas. The first is research, which means studying how advanced AI models learn and behave. The second is product development, which turns that research into tools like Claude, Claude Code, and Cowork. The third is safety, which means testing models and finding ways to reduce risk.
This mix matters. Anthropic does not rebrand someone else’s chatbot. It trains its own models and builds the products around them.
Who Founded Anthropic?
Anthropic was founded in 2021 by a group of AI researchers and tech leaders. Many of them had worked at OpenAI. Dario Amodei is the co-founder and CEO, and his sister Daniela Amodei is the co-founder and president.
The other well-known co-founders are Tom Brown, Jared Kaplan, Jack Clark, Sam McCandlish, and Chris Olah. Brown was involved in the company’s early technical work. Kaplan is known for research on how AI improves as models get larger. Clark has worked on AI policy, and Olah is known for research into how neural networks work inside.
Why Was Anthropic Created?
The founders believed that making AI more powerful was only half the job. Researchers also needed better ways to understand model behavior, test for risks, and guide AI toward helpful answers. Anthropic was built around that idea.
A powerful AI model can give wrong answers, follow harmful instructions, or act in ways its builders did not expect. These problems matter more as AI gets access to software, business data, and outside tools. The company’s stated purpose is to develop and maintain advanced AI for the long-term benefit of humanity.
Is Anthropic a Nonprofit?
No. Anthropic is a for-profit Public Benefit Corporation (PBC). It sells products, earns revenue, and raises money from investors like any other business.
What makes a PBC different is its legal duty. It must weigh its stated public benefit purpose alongside the financial interests of its shareholders. Anthropic also has a Long-Term Benefit Trust, which has a formal role in company governance and board appointments. The trust is meant to protect the company’s mission as it grows, but it does not remove the pressure to earn money.
What Does Anthropic Do?
Anthropic does five main things. It builds AI models, researches how they work, works on AI safety, sells AI to businesses and developers, and studies how AI affects society. Each one is covered below.
It Builds the Claude AI Models
Anthropic develops Claude, a family of large language models. These models can read and write text, analyze files, write code, and use tools. They handle everything from simple questions to long, multi-step work projects.
A person might ask Claude to explain a hard topic, edit a document, or summarize a long report. A developer might ask it to find a bug. Businesses also build their own apps on top of Claude.
It Does Advanced AI Research
Anthropic has research teams that study what AI models can do and where they fall short. They look at how models learn, how they react to different instructions, and what happens when they face risky tasks. This goes well beyond making Claude answer questions better.
The company names several research areas. Alignment focuses on guiding model behavior. Interpretability studies how models work internally. Frontier red teaming tests advanced models for risks in cybersecurity, biosecurity, and autonomous systems. Economics and societal impacts research looks at how AI changes jobs, productivity, and daily life.
It Works on AI Safety
Safety is a core part of Anthropic’s research. In this context, AI alignment means trying to make a model follow human instructions, goals, and safety expectations. The company uses several methods to get there.
One of Anthropic’s best-known safety approaches is Constitutional AI, a training method that uses written principles to help models critique and improve their responses. Anthropic also publishes Claude’s Constitution, a document describing the values and behavior it wants Claude to follow. The company uses model evaluations, red teaming, safety classifiers, and interpretability research alongside these approaches. These methods can reduce risk, but they cannot eliminate it. Claude can still make mistakes.
It Sells AI to Businesses and Developers
Businesses use Claude to analyze documents, support software teams, search internal information, and prepare reports. Developers can plug Claude into their own apps through the Claude API. Anthropic has also built tools and integrations for fields like financial services, healthcare, and small business.
Anthropic supplies the technology. Its customers decide which tasks to hand to AI and how much human review those tasks need.
It Studies How AI Affects Society and the Economy
Anthropic also studies the wider effects of AI. Its economics research asks how AI could change work, productivity, and opportunity. Its societal impacts research looks at how people and organizations actually use AI. This matters because AI can change how companies hire, how people get work done, and how information spreads.
What Products Does Anthropic Offer?
Most of Anthropic’s customer-facing products are built around Claude. The company also develops technologies and open standards, such as MCP, that can be used by other AI systems.
Claude
Claude is Anthropic’s main AI assistant. People use it on the web and in mobile and desktop apps. It helps with writing, research, explanations, document analysis, and coding. A small-business owner might upload a report and ask for the key points, while a student might ask for help with a tough topic. Claude is similar to ChatGPT in what it does, but ChatGPT comes from OpenAI.
Claude Code
Claude Code is Anthropic’s AI coding agent. It does more than suggest lines of code. It can work inside a software project, read files, make changes across several files, run commands, and help test the results.
For example, a developer could ask it to look into an error in a website’s login system. Claude Code can inspect the files and propose and test a fix. Developers still need to review the changes, because AI-written code can contain mistakes or security gaps.
Claude Cowork
Claude Cowork is built for knowledge work that involves files, tools, and many steps. Instead of just answering a question, it can carry out a larger task using the files and apps a user allows it to use. Examples include comparing several documents, organizing research, or preparing a report.
It is aimed at analysts, researchers, project managers, and other people who work with digital information all day. Because Cowork can touch files and connected tools, users should check its permissions and review important results.
Claude API and Developer Platform
The Claude API lets developers use Anthropic’s models inside their own websites, software, and customer-service systems. The developer platform also offers tools for building AI agents. A company could use it to build a support assistant, a document-analysis tool, or an internal search system. Anthropic generally charges for API use based on how much text goes into and comes out of the model.
Claude for Business and Enterprise
Anthropic offers business and enterprise plans for companies that want Claude across many teams. These plans are built for workplaces that need team management, admin controls, connected tools, and support. Common uses include software engineering, research, document analysis, and customer support.
Anthropic also offers industry-focused products. Its announcements cover small businesses, financial services, healthcare and life sciences, nonprofits, and creative professionals. These tools help professionals, but they do not replace qualified financial, medical, or legal experts.
Model Context Protocol (MCP)
Anthropic introduced the Model Context Protocol, or MCP, in 2024. MCP is an open standard that helps AI apps connect to outside tools and data, such as file systems, databases, and business software. Without a shared standard, developers would need to build a separate connection for each app.
Anthropic later donated MCP to the Agentic AI Foundation under the Linux Foundation. MCP is not a Claude plan. It is a technology that many different AI apps can use.
What Does Anthropic Do Besides Claude?
Claude is the face of the company, but much of Anthropic’s work happens behind the scenes. Two areas stand out.
The first is mechanistic interpretability. Large language models are built from huge networks of math, and their answers are hard to explain just by reading them. Anthropic’s researchers study how information moves through a model so they can understand why it behaves the way it does. This field is still young, and today’s tools cannot explain everything a large model does.
The second is risk research and red teaming. The Frontier Red Team tests models for serious risks in areas like cybersecurity and biosecurity. A model can act differently under unusual or hostile instructions than it does in normal use, so this testing matters. The results help shape safeguards, access rules, and product decisions.
Claude is also used in science and medicine. Researchers can use it to read papers, organize findings, and analyze data. Claude does not discover or approve medicines on its own, and scientific results still need expert review and proper testing.
How Does Anthropic Make Money?
Anthropic makes money by selling access to its AI models and products. Its customers include individual subscribers, developers, startups, and large companies. According to the company’s SEC filing, its revenue comes in two main forms.
The first is usage-based revenue. This is money earned when customers use Anthropic’s models, mostly through the API. A company that sends millions of requests to Claude pays more than a developer running a small test project.
The second is subscription revenue. Individuals can pay for Claude plans that give higher usage limits and more features than the free plan. Businesses can also buy Claude for their teams, and some large customers sign annual or multi-year spending agreements.
Anthropic also sells through major cloud providers, including Amazon Web Services, Google Cloud, and Microsoft. This lets customers reach Claude through platforms they already use. These partners also supply some of the computing power that advanced AI needs.
How Much Revenue Does Anthropic Make?
Anthropic’s October 2026 registration filing with the U.S. Securities and Exchange Commission (SEC) reported the following revenue.
| Period | Reported revenue |
|---|---|
| 2023 | $41 million |
| 2024 | $386 million |
| 2025 | $4.585 billion |
| First half of 2025 | $1.163 billion |
| First half of 2026 | $16.286 billion |
The first-half rows cover six months, while the other rows cover full years, so they are not equal periods. The filing also shows that about 80% of revenue in the first six months of 2026 was usage-based and about 20% was subscription-based. In other words, developers and business customers drive most of Anthropic’s income, not individual Claude subscribers.
Is Anthropic Profitable?
Not yet, based on its filing. Anthropic reported operating losses in 2023, 2024, 2025, and the first half of 2026. Its operating loss in the first half of 2026 was about $693 million, down from about $2.14 billion in the first half of 2025.
Running advanced AI is expensive. The filing reports $8.945 billion in compute and infrastructure costs for the first half of 2026. The filing also lists large accounting items outside of operations, so operating loss and net loss are not the same number.
How Does Anthropic Handle User Data?
People share personal details, work files, and code with Claude, so privacy rules matter. Anthropic has different policies for consumer products and commercial services, so the answer depends on which one you use.
Does Anthropic Train Its Models on Your Chats?
For consumer plans like Claude Free, Pro, and Max, Anthropic may use chats and coding sessions to improve its models when the user allows it. The policy lists exceptions, such as conversations flagged for safety review. For commercial products like the API and Claude for Work, Anthropic says it does not train on customer inputs and outputs by default unless the customer gives permission or another stated exception applies.
How Long Does Anthropic Keep Your Data?
On consumer accounts, you can delete a conversation from your history. Anthropic says deleted chats are removed from its backend within 30 days under its standard policy. There are exceptions, though. Data tied to certain policy violations or to feedback you submit may be kept longer, and some de-identified data may be kept for research or statistics.
Commercial products have their own data retention rules. Anthropic’s standard API does not retain most prompts and responses by default, although some features and models may have a 30-day retention period. Eligible customers can also request zero data retention for supported features. Cowork and Claude Code may follow separate retention policies, so businesses should check the terms for the exact product they use.
How Can You Protect Your Information?
A few habits help. Check the privacy and model-improvement settings on your account. Avoid sharing passwords or extra personal details. Review the permissions before you connect outside apps. For confidential work files, use a commercial plan and follow your employer’s data rules.
What Does Anthropic Do for the U.S. Military?
Anthropic works with U.S. government and national security organizations. According to the company, Claude has been deployed on classified government networks. Anthropic describes uses such as intelligence analysis, modeling and simulation, operational planning, and cyber operations. So the company’s customers include government agencies as well as consumers and private businesses.
Why Did Anthropic Clash With the Department of War?
The disagreement centered on two uses of AI that Anthropic wanted to limit. The first was mass domestic surveillance of people in the United States. The second was fully autonomous weapons, because Anthropic argued that today’s AI is not reliable enough to make lethal targeting decisions without human control. The company said it supported lawful national security uses outside those two limits.
The Department of War wanted broader permission to use AI for any lawful military purpose. The two sides did not reach a deal.
What Happened Next?
The dispute led to government action and two separate court cases in 2026. In February 2026, Anthropic publicly described the disagreement, and the Secretary of War announced action to designate the company a supply-chain risk. In March, federal agencies were directed to stop using Anthropic’s technology, and the company challenged the actions in court.
In August 2026, a federal court in California set aside the challenged government actions in its case. But in September 2026, a separate federal appeals court in Washington, D.C., upheld another supply-chain-risk designation under a different federal law. The two rulings had different legal effects. The California decision did not cancel the separate designation upheld by the D.C. Circuit, meaning the Department of War could continue efforts to remove Anthropic products from its own systems under that authority.
These are two different cases, and one ruling does not cancel the other. The larger question is who should decide how advanced AI is used in national security. Anthropic says some uses should stay off limits. Government officials argue for broader access for lawful missions.
What Are the Main Controversies Around Anthropic?
Anthropic presents safety as central to its mission, but it has also faced criticism and lawsuits. The biggest issues involve copyright, safety commitments, military use, and cost.
Copyright and Training Data
A group of authors sued Anthropic, saying it used copyrighted books without permission to train Claude. The case raised two separate questions. One was whether training on lawfully obtained books can count as fair use. The other was whether keeping pirated copies of books is infringement. The court found that training on lawfully obtained books could qualify as fair use, but that storing pirated copies was a separate problem.
In July 2026, a judge gave final approval to a $1.5 billion class settlement of the authors’ claims. This was a major moment for the AI industry. It does not mean all AI training on copyrighted work is legal, and it does not settle every copyright dispute involving Anthropic.
Questions About Safety
Even with heavy spending on safety research, hard questions remain. Can evaluations catch every dangerous ability? Will models always follow their safeguards? How much can researchers really see inside a model?
In October 2026, Anthropic updated its safety roadmap. It retired one goal, a routine review of a sample of training data and rewards, after finding the process did not catch problems beyond those that existing reviews already found. The company said it would keep updating Claude’s public Constitution, running automated alignment assessments, doing interpretability research, and publishing results in system cards and risk reports. That was a change to one process, not an end to safety research.
Military Limits and the Cost of AI
People disagree about the dispute with the Department of War. Some back Anthropic’s limits on mass surveillance and autonomous weapons. Others say a private company should not restrict the government’s access to tools for lawful national security work.
Cost is another challenge. Building and running advanced AI takes huge amounts of computing power, energy, and skilled staff. Anthropic’s filing shows how large these costs are, even as revenue grows quickly.
How Is Anthropic Different From OpenAI and Google DeepMind?
Anthropic competes with OpenAI and Google DeepMind. All three build advanced AI models and sell products to consumers, developers, and businesses. Their products and company structures differ.
| Feature | Anthropic | OpenAI | Google DeepMind |
|---|---|---|---|
| Main AI family | Claude | GPT | Gemini |
| Main assistant | Claude | ChatGPT | Gemini |
| Coding product | Claude Code | Codex | Gemini coding tools |
| Developer access | Claude API | OpenAI API | Gemini API |
| Business offering | Claude for Business and Enterprise | ChatGPT business and enterprise plans | Google Cloud AI services |
This table compares products, not quality. It is not a ranking of which AI is best.
What sets Anthropic apart is the mix of products, a stated public benefit mission, and research into model behavior. Its work on Constitutional AI, interpretability, and frontier risk testing is easy to see. Its Public Benefit Corporation status and Long-Term Benefit Trust also make its governance different from most tech companies.
How Widely Do Businesses Use Anthropic?
Business customers are a big part of Anthropic’s growth. In its SEC filing, the company reported figures as of September 30, 2026. It had more than 11,900 customers that each generated at least $100,000 in revenue over the past 12 months. More than 1,900 customers generated at least $1 million, and about 250 generated at least $10 million.
These numbers show that Anthropic serves more than individual users and small developer teams. A software company might use Claude Code for engineering work, while another organization might use the API to power an internal research tool. Anthropic also describes AI that can work across approved tools and company data as a kind of virtual collaborator. The goal is to move AI from answering questions to helping finish real work.
Frequently Asked Questions About Anthropic
What Does Anthropic Do?
Anthropic builds advanced AI models, makes products like Claude and Claude Code, and researches how to make AI safer. It sells its technology to individuals, developers, businesses, and government agencies.
What Is Anthropic Best Known For?
Anthropic is best known for Claude. It is also known for AI safety research, Claude Code, enterprise AI services, and its work on how large models function inside.
Is Anthropic the Same as Claude?
No. Anthropic is the company, and Claude is its family of AI models and products.
Is Anthropic a Nonprofit?
No. Anthropic is a for-profit Public Benefit Corporation. It earns revenue and raises investment, and it must also weigh its stated public benefit purpose.
Who Owns Anthropic?
Shareholders own Anthropic. That group includes its co-founders, investors, and other stockholders. The Long-Term Benefit Trust also has a formal governance role, including authority over some board appointments.
Is Anthropic Publicly Traded?
Not based on the filing alone. Anthropic filed a Form S-1 with the SEC on October 26, 2026, for a proposed initial public offering. The filing names a proposed Nasdaq listing under the ticker ANT. A filing does not mean shares are trading, so check recent news for the current status.
Is Claude Safe to Use?
Claude has safeguards meant to reduce harmful answers, and Anthropic tests its models. No AI assistant is risk-free, though. Claude can give wrong information or misread instructions, so check important answers and review the privacy rules for your account before sharing sensitive details.
Is Claude Free?
Anthropic offers a free Claude plan and paid plans with more features and higher usage limits. Developers and businesses can also pay to use Claude through the API and commercial services.
Who Are Anthropic’s Main Competitors?
OpenAI and Google DeepMind are the main ones. Anthropic also competes with other companies that build AI models, coding agents, developer platforms, and enterprise AI products.
The Bottom Line: What Does Anthropic Do?
Anthropic is an AI research and technology company. It builds the models, tools, and services that let people and organizations use advanced AI. Claude is its most recognized product, but it is only one part of the company’s work.
The business runs mostly on usage-based AI services, along with subscriptions and enterprise agreements. Its public benefit structure gives its stated mission a formal place in company governance. Because details like financial results, privacy terms, legal cases, and IPO status can change, check Anthropic’s official pages and its SEC filings for the latest information.
You Might Also Like:
What Is CDiPhone? Meaning, Features, and the Truth Behind the Name
What Is an Internet Provider? How an ISP Works
I’m Emma Rose, the founder of Punstation. I cover technology, digital trends, business, and online tools, creating practical guides and thoughtful reviews to help readers understand their options and make informed decisions. For more on how I approach my content, see our Editorial Standards.