Anthropic: The Company Behind Claude
Anthropic is the company behind Claude. Understanding it means separating three questions: what its tools do, how access is sold, and what evidence supports its claims about safety and public benefit.
▶ Explore the narrated 3D explainerKey takeaways
- Anthropic is the company; Claude names models and products, available through different interfaces and plans.
- A public benefit purpose, a constitution and a risk policy describe governance and intentions. Published evidence remains essential.
- October’s announcements have distinct scopes: evaluator access, a conditional equity commitment and an open standard.
What does Anthropic build?
Anthropic: Company and governance describes an AI research and product company structured as a public benefit corporation. Claude: Products and pricing lists apps, coding tools and developer access. The catalogue currently includes the Fable, Opus, Sonnet and Haiku model families. Those are model labels, rather than subscription plans or an independent performance ranking.
Different interfaces put capabilities into different workflows. A chat application, a coding tool and an API are not interchangeable products. Availability and limits depend on the offering. The explainer’s connecting blocks are an illustrative product diagram, not Claude’s architecture, actual traffic or private customer information.
How is access sold?
The official catalogue includes free access, paid subscriptions and business plans. Developer pricing distinguishes input from output tokens and can include additional feature charges. Check the current terms for the specific product and model. A catalogue describes what customers can buy; it does not establish audited revenue, customer totals or profitability.
What does the safety approach mean?
Anthropic: Claude’s Constitution sets out intended values and behaviour. Anthropic acknowledges that models may not always reflect those ideals. Anthropic: Responsible Scaling Policy addresses risk governance and reporting; its page was updated on 14 August 2026 and identifies version 3.4 as effective on 8 July. A behavioural guide and an organisational risk policy serve different purposes. Neither is proof that every response or deployment is safe.
The company says its board is elected by stockholders and the Long-Term Benefit Trust. That corporate structure is one accountability mechanism. It should be considered alongside evaluations, incident reporting and evidence of how mitigations perform.
What changed in October?
| Announcement date, 2026 | Company statement | Scope or condition | Primary source |
|---|---|---|---|
| 15 October; clarified 16 October | Deeper METR evaluator access | Stated access and publication terms; not a blanket certification | Anthropic: Independent evaluators update |
| 18 October | 5% of company shares committed to a Public Benefit Fund | Subject to stockholder approval; equity, not cash already distributed | Anthropic: Public Benefit Fund |
| 22 October | Agent Skills donation announced | Open standard and related assets, not users’ individual skills | Anthropic: Donating Agent Skills |
All three rows describe company announcements checked on 24 October 2026. They are different kinds of commitments and should not be combined into a single measure of safety or public benefit.
The evaluator update reports deeper access for METR and discusses publication terms, including specified redactions. The practical questions are what evaluators can inspect, what they can publish and what their findings show. This page does not treat access itself as a safety verdict.
The fund announcement concerns company equity. Its stockholder-approval condition remains attached here: we do not assert that approval has occurred or that a matching amount of cash has been distributed. The Agent Skills announcement concerns an open standard’s institutional home at the Agentic AI Foundation; it does not transfer ownership of everyone’s individual skills.
Frequently asked questions
What is the difference between Anthropic and Claude?
Anthropic is the company. Claude names its AI models and products. Its apps, coding tools and developer platform are different ways to access capabilities, with availability depending on the product and plan.
How does Anthropic charge for Claude?
The official catalogue includes free access, paid subscriptions and business plans. Developer pricing includes input and output tokens, with other charges for some features. A pricing catalogue does not establish revenue, customer counts or profitability.
Does a public benefit structure guarantee safe AI?
No. Corporate purpose, behavioural guidance and risk policies describe intentions and governance. They do not establish that every model response or deployment is safe. Published evaluations and the scope of external scrutiny still matter.
Did Anthropic give away 5% of its cash?
No. Its 18 October 2026 announcement described a commitment of 5% of company shares to a Public Benefit Fund, subject to stockholder approval. Equity is not cash already distributed, and this page does not assert that approval has occurred.
What was the Agent Skills announcement?
On 22 October 2026, Anthropic announced donating the open Agent Skills standard to the Agentic AI Foundation. The announcement concerns the standard and related assets, not ownership of users’ individual skills.
Sources and reporting cutoff
- Anthropic: Company and governance. Company description and governance, checked 24 October 2026.
- Claude: Products and pricing. Current product catalogue, model families and billing descriptions, checked 24 October 2026. Provider descriptions, not independent performance findings.
- Anthropic: Claude’s Constitution. Intended values and behaviour; the company acknowledges that models may not always reflect these ideals. Checked 24 October 2026.
- Anthropic: Responsible Scaling Policy. Risk governance and reporting; page updated 14 August 2026, version 3.4 effective 8 July 2026. Checked 24 October 2026.
- Anthropic: Independent evaluators update. Published 15 October 2026; clarified 16 October. Reports deeper METR access, subject to stated scope and publication terms. Checked 24 October 2026.
- Anthropic: Public Benefit Fund. Published 18 October 2026. A commitment of 5% of company shares, subject to stockholder approval; not cash already distributed. Checked 24 October 2026.
- Anthropic: Donating Agent Skills. Published 22 October 2026. Announced donation of the open Agent Skills standard to the Agentic AI Foundation; not ownership of users’ individual skills. Checked 24 October 2026.
Research checked 24 October 2026. These primary sources establish what the provider describes or announces. They are not independent verification of all outcomes. Recheck product terms, policies and published findings before a later render.
▶ Explore the seven scenesCharts, maps and diagrams from the explainer






