This evergreen history article uses authoritative archives and official records. Exact dates are used when documented; gradual inventions and rollouts are described as periods rather than being assigned a misleading single birthday.
Quick facts
- Anthropic was founded in 2021 in San Francisco by former OpenAI researchers and leaders.
- The company focuses heavily on reliable, interpretable and controllable AI systems.
- Claude was introduced publicly in March 2023.
- Constitutional AI uses written principles as part of a model-training and evaluation process.
- Anthropic distributes Claude through consumer, developer and enterprise channels.
Founding after the first large-model boom
Anthropic was created in 2021 by a group that included siblings Dario Amodei and Daniela Amodei, along with other former OpenAI employees. The founders argued that rapid progress in large language models made safety, alignment and interpretability central engineering problems rather than distant theoretical concerns.
The company established itself in San Francisco and attracted investment from major technology partners. Its research agenda included scalable oversight, model behaviour, interpretability and methods for reducing harmful outputs while preserving broad usefulness.
Constitutional AI
A distinctive Anthropic research direction was Constitutional AI. Instead of relying only on humans to label every preferred answer, the method asks a model to critique and revise outputs according to a written set of principles. Human feedback remains important, but the constitution provides an explicit framework for model behaviour.
The approach was presented as a way to make training more transparent and scalable. It also raised practical questions: who chooses the principles, how conflicts are resolved and how a model should behave across cultures, legal systems and high-risk domains. Anthropic has continued revising and publishing material about Claude’s intended values.
The launch of Claude
Anthropic introduced Claude on 14 March 2023. The first public offering included Claude and a faster Claude Instant model through chat and API access. The assistant was designed for conversation, summarisation, analysis, writing and other language tasks, with emphasis on predictable and helpful behaviour.
Claude entered a fast-growing market shaped by ChatGPT, Google’s models and open-source alternatives. Anthropic differentiated itself through long context windows, enterprise partnerships, safety research and a restrained product identity. Developers used Claude for document analysis, coding and tool-connected workflows.
Model families, coding and enterprise use
Later Claude generations improved reasoning, vision, coding and tool use. Anthropic expanded access through its own platform and partnerships with cloud providers. Claude became prominent in software development because it could work with large codebases, produce structured edits and participate in longer multi-step tasks.
Enterprise adoption brought security, privacy and reliability requirements. Model providers had to address data retention, tenant isolation, prompt injection, tool permissions and auditability. Anthropic’s product history therefore combined model research with the less visible work of building trustworthy infrastructure.
Competition and the future of controllable AI
Anthropic’s rise demonstrated that the frontier-model market could support more than one major laboratory. Competition encouraged faster capability gains but also increased spending on chips, data centres and talent. The company faced the same unresolved problems as its rivals: hallucination, misuse, copyright, concentration of power and uncertain evaluation of advanced systems.
Claude’s longer-term significance may depend less on benchmark rankings than on whether constitutional training, interpretability and safety engineering translate into dependable real-world behaviour. Anthropic’s history is still short, but it has already influenced how the industry discusses model character, oversight and responsible deployment.
Common misconceptions
- Anthropic was not founded in the nineteenth century despite the classical-sounding name; it was created in 2021.
- Constitutional AI does not mean that Claude is governed by a national constitution.
- A published set of principles does not eliminate model errors or harmful outputs.
- Claude is a family of models and products, not one permanently fixed system.
Timeline: key years and locations
| Year | Location | Event | Why it mattered |
|---|---|---|---|
| 2021 | San Francisco, California, United States | Anthropic is founded | Created a frontier AI company centred on safety and reliability. |
| 2022 | San Francisco research community | Constitutional AI research is published | Introduced principle-guided critique and revision in model training. |
| March 2023 | Online launch from San Francisco | Claude and Claude Instant are introduced | Brought Anthropic’s assistant to businesses and developers. |
| 2023 | United States cloud market | Major cloud partnerships expand | Provided capital, compute and enterprise distribution. |
| 2024 | Global | Claude 3 family broadens capability | Strengthened reasoning, vision and enterprise adoption. |
| 2025 onward | Global developer ecosystem | Coding, tool use and agent workflows grow | Moved Claude from chat toward longer operational tasks. |
Frequently asked questions
Who founded Anthropic?
Anthropic was founded by a group of former OpenAI employees led by Dario Amodei and Daniela Amodei.
When was Claude launched?
Anthropic introduced Claude publicly on 14 March 2023.
What is Constitutional AI?
It is a training approach in which models critique and revise responses using an explicit set of written principles, alongside other feedback and safety methods.