Anthropic is an AI company based in San Francisco that builds a family of AI models called Claude. It was founded in 2021 by a group of former OpenAI researchers and staff, and it describes its mission as developing AI that is safe and useful for people. If you have used a chatbot called Claude, you have used Anthropic’s product.
That is the short answer to what Anthropic is. The longer answer is a story with some well-documented parts and some parts that are still unsettled, and it helps to know which is which. This post walks through it one question at a time, then ends on what it all means if you write or run a newsletter.
So what is Anthropic, in one coffee-sized answer?
Anthropic is a company that trains large language models, the same kind of technology behind ChatGPT and Google’s Gemini. You type or paste text, the model reads it and writes a response. Anthropic’s version is Claude.
Two things set the company’s identity apart from many of its competitors. It is organized as a public benefit corporation, a legal structure that lets a company weigh a stated public mission alongside profit. And it publishes a good deal of research about AI risk and about how to make models behave reliably.
Whether that identity changes what you experience as a user is a fair question, and we will get to it.
Where it came from: the people and the split that started it
Anthropic was founded in 2021 by a group that had worked at OpenAI. The best-known names are the siblings Dario Amodei, who is the CEO and had been a research leader at OpenAI, and Daniela Amodei, who is the president. Several other senior OpenAI researchers joined them as co-founders.
Why did they leave? The founders have spoken publicly about wanting to put safety research at the center of how AI is built. News coverage at the time also pointed to disagreements over direction and priorities at OpenAI, including how quickly to commercialize. Nobody outside the room can fully reconstruct those conversations, and accounts differ in emphasis. What is solid is the outcome: a group of experienced researchers left and started a rival lab with safety as its stated founding purpose.
It is worth keeping that in proportion. A company’s own description of its motives is a claim, and a reasonable one to hold lightly. The more useful test is what the company does over time, which the next sections cover.
What Anthropic actually makes, and what Claude is
The product is Claude, a family of AI models. Anthropic has released them in generations, and within a generation it has usually offered models at different sizes, with names such as Haiku (smaller and faster), Sonnet (the middle option) and Opus (the most capable). Specific model names and versions change often, so check Anthropic’s own site for the current lineup rather than trusting any article, this one included.
You can reach Claude in a few ways:
- The Claude app and website. You chat with it directly. There is a free tier and paid plans for individuals, teams and larger organizations.
- The API. Developers and companies plug Claude into their own software, such as customer support tools or writing assistants.
- Cloud platforms. Claude is also offered through large cloud providers, including Amazon’s and Google’s, so businesses can use it inside systems they already run.
- Coding tools. Anthropic makes a tool called Claude Code, aimed at software developers.
In practice, people use Claude for drafting and editing text, summarizing long documents, answering questions about material you paste in, brainstorming, and writing code. Like every current language model, it can state wrong things confidently, so anything factual needs checking before it goes in front of readers.
Why “safety” keeps coming up when people talk about this lab
Safety is the word most associated with Anthropic, and it covers a few concrete things rather than one vague promise.
Constitutional AI. This is a training method Anthropic developed. Instead of relying only on people rating thousands of responses, the model is given a written set of principles and learns to critique and revise its own answers against them. Anthropic has published papers on the approach and has released versions of the principles it uses for Claude.
A Responsible Scaling Policy. Anthropic published a framework that ties the precautions it takes to how capable its models are. The idea is that the more dangerous a model could be in the wrong hands, the stricter the testing and security should become before it is released. It is a company’s own commitment, not a law, and the company has revised the policy over time.
Interpretability research. Anthropic has a research effort aimed at understanding what is happening inside a model, so that behavior can be explained instead of only observed.
Governance. As a public benefit corporation, Anthropic has a legal basis for weighing its mission. It also created a Long-Term Benefit Trust, a body with the power to elect some members of the company’s board, intended to keep long-term public interest in view.
What remains unsettled is how much any of this guarantees. Critics point out that a company building frontier AI while warning about AI risk has an obvious tension, and that voluntary policies are only as strong as the company’s willingness to follow them. Supporters respond that it is better to have safety-minded labs at the frontier than to leave it to others. Both arguments are live, and reasonable people land in different places.
How Anthropic pays its bills: the business behind the research
Training a frontier model requires enormous amounts of computing power, and that costs a great deal of money. Safety research does not pay for itself, so Anthropic needs revenue and investors.
Its income comes mainly from two places: businesses that pay to use Claude through the API and cloud platforms, and people and organizations that pay for Claude subscriptions. Software development has been a significant use, which is one reason Anthropic built tools aimed at programmers.
On the funding side, Anthropic has raised large investments, including from Amazon and Google, both of which also supply computing infrastructure. That matters for a plain reason: a lab that depends on big technology partners for money and hardware has commercial pressures like any other company. Exact valuations and funding totals move quickly and get reported differently by different outlets, so I am leaving specific figures out rather than printing numbers that may be stale by the time you read this.
The honest summary is that Anthropic is both a research lab and a commercial product company, and the two goals do not always point the same way. Watching how it handles the moments where they conflict tells you more than its mission statement does.
Why people are watching it now, and what is still uncertain
Anthropic gets attention for a few reasons. Claude is widely regarded as one of the leading AI models, particularly for writing and coding. The company has become one of the best-funded AI labs. And it has taken public positions on how AI should be regulated, which puts it in policy conversations as well as product ones.
Several things remain open:
- Competition. OpenAI, Google, Meta and others release new models constantly. A lead in one generation can vanish in the next.
- Cost. Building and running these models is expensive, and it is not settled how the economics will work out for any lab over the long run.
- Legal questions. AI companies, Anthropic included, have faced lawsuits from authors and other rights holders over the material used to train models. Courts and lawmakers are still working out the rules.
- Safety claims. Whether the safety commitments hold up as models become more capable is something only time and outside scrutiny can answer.
If someone tells you with total certainty where any of this ends up, they are guessing.
The Anthropic IPO
Anthropic is scheduled to price its initial public offering on the evening of Wednesday, November 18, 2026, and begin trading on the Nasdaq Global Select Market under the ticker symbol “ANT” on Thursday, November 19, 2026
Price Range: $70.00 to $78.00 per share [1]
Target Valuation: Roughly $1.3 trillion to $1.45 trillion (based on a targeted raise of up to $64.4 billion for 825 million offered shares
When you are reading this post IPO the business will be very different
What this means if you write or run a newsletter
For a newsletter writer, Anthropic matters mostly as the maker of a tool, and a tool can be evaluated on practical grounds.
It can speed up the unglamorous parts. Summarizing a long report, tightening a draft, suggesting subject line options, turning interview notes into an outline: these are tasks where a model like Claude can save time. You still decide what is worth saying.
You are responsible for accuracy. A language model can invent a quote, a statistic or a source. If your newsletter reports facts, verify each one against the original before you send. Your readers trust your name, not the model’s.
Your voice is the part that stays yours. Newsletters survive on a reader’s sense that a particular person is on the other end. Using AI for first-pass drafting and then rewriting in your own words and judgment keeps that intact. Sending unedited model output tends to produce the generic tone readers learn to skim past.
Read the terms before you paste anything sensitive. If you handle subscriber data, private emails or unpublished work, check the current privacy and data-use terms of whichever AI tool you use, since they differ by plan and change over time.
Your tool choices can change. Because the leading models trade places, it makes sense to treat any single AI tool as something you can swap, not a foundation your newsletter depends on.
None of this settles the bigger question of whether a newsletter business still makes sense in a world where AI can produce text on demand. That is a question about audience, trust, and what readers pay for, and I take it on in a separate post: do newsletters still work when AI can write everything? It is a good next read if the rise of labs like Anthropic has you wondering where that leaves your own writing.
After that, you can look through the newsletter writing and growth tools recommended on this site. They are options to try, not things you have to buy, and the sensible way to choose is to test one against a real issue of your newsletter and keep it only if it saves you time or helps you reach readers.
Post by Peter Hanley
November 2026
Anthropic