Anthropic and Claude: how a safety-first AI lab built a category position alongside OpenAI
Anthropic was founded in 2021 by Dario and Daniela Amodei and a group of former OpenAI researchers concerned about the trajectory of frontier AI development. The company was structured as a Public Benefit Corporation with an explicit AI-safety mission. Anthropic completed training of its first Claude model in 2022, ahead of ChatGPT's public release, and made Claude broadly available in 2023. The company's positioning has been built around Constitutional AI (a technique that trains models against a written constitution rather than relying entirely on human feedback) and around public-policy and safety-research credibility with regulators and enterprise buyers.
- Story: Anthropic founded 2021 by former OpenAI researchers Dario and Daniela Amodei with AI safety positioning. Launched Claude AI assistant 2023. Claude 3 (March 2024) and Claude 3.5 (June 2024) capable competitor to ChatGPT. Major investors include Amazon (~$4B+) and Google. Enterprise adoption particularly in regulated industries.
- Why it matters: Anthropic is a defining AI safety lab category emergence case — demonstrating safety positioning differentiation in AI lab competition.
- Takeaway: Differentiation through safety positioning produces enterprise customer preference particularly in regulated industries.
- Takeaway: Major-investor strategic relationships can fund infrastructure-intensive AI development.
- Takeaway: AI lab competition is genuine despite small number of major competitors.
Anthropic Claude — the four-step story
Anthropic by the numbers
Quick facts
Where AI was in 2021
In 2021, large-language-model research was concentrated at a small number of organisations: OpenAI (which had released GPT-3 in 2020), DeepMind (Google), and Meta AI. The frontier capability of these models was advancing rapidly, but commercial deployment, safety research, and public-policy frameworks for the technology were lagging. A group of researchers at OpenAI, including the Amodei siblings, became concerned about the pace of capability development relative to safety research. They left to found Anthropic with the explicit intention of running an AI lab where safety research would be load-bearing rather than secondary.
The founding bet was specific: build a frontier AI lab with safety research and frontier model development under the same roof, on the theory that the two needed to inform each other to be effective. The public-benefit-corporation structure was chosen to signal that the safety mission was structurally protected from pure profit-maximisation pressure.
The Claude product and Constitutional AI
Anthropic completed training of Claude 1 in 2022, ahead of OpenAI's November 2022 release of ChatGPT. Claude was initially made available in limited-access settings while Anthropic worked on safety evaluation and on the techniques that would shape the public product. A broader release came in 2023, by which point ChatGPT had created enormous public awareness of conversational AI.
The technical and brand differentiator was Constitutional AI (CAI). Rather than rely entirely on human-feedback loops to align model behaviour, CAI uses a written set of principles — a constitution — that the model is trained to follow. The constitution draws from sources including the UN Universal Declaration of Human Rights and AI-safety research principles. The technique was published in Anthropic research papers and refined across subsequent Claude model generations. Claude's constitution was publicly released in 2025 as part of the company's ongoing transparency around alignment methodology.
How the positioning has played out
By 2024-2025, Anthropic had established a meaningful market position alongside OpenAI in frontier model development and commercial deployment. Major cloud-provider partnerships with Amazon (AWS) and Google made Claude available to enterprise customers through the channels they already used. Anthropic published research on AI safety, interpretability, alignment, and frontier-model risks at a cadence comparable to academic AI labs, building public-policy and regulatory credibility that contrasted with the more product-focused communications of some competitors.
The strategic bet has been that enterprise customers and regulators would value the safety-first positioning enough to choose Anthropic over alternatives even when product capability was comparable. The bet has paid off in regulated industries (financial services, healthcare, legal) and with public-sector buyers. The competitive question going forward is whether the positioning continues to translate into customer wins as frontier capabilities continue to advance across the industry.
How RGM thinks about category positioning
When clients ask about positioning in a category where competitors have head-start scale (Anthropic vs OpenAI being one current example), the Anthropic playbook is instructive. Three structural choices. First, pick a positioning attribute that the leader cannot easily copy without contradicting their existing brand — OpenAI's positioning could not credibly pivot to safety-first without contradicting prior public commitments. Second, back the positioning with structural commitments (PBC status, published safety research, transparency around alignment methodology) so that buyers and regulators see consistency over time. Third, build the channel strategy around customer segments where the positioning attribute is most load-bearing — regulated industries valued safety-first more than consumer apps did.
The pattern is hard to copy when the positioning attribute is not load-bearing for any specific customer segment, or when the structural commitments are easily reversed. Both Anthropic's PBC structure and its public safety research are difficult to walk back without significant reputational cost, which is what gives the positioning long-term durability.
Frequently asked questions
When was Anthropic founded?
2021 by Dario and Daniela Amodei and a group of former OpenAI researchers. The Amodei siblings serve as CEO and President respectively. Dario was previously VP of Research at OpenAI.
What is Constitutional AI?
A training technique developed by Anthropic that aligns AI models against a written constitution of principles rather than relying entirely on human-feedback loops. The constitution draws from sources including the UN Universal Declaration of Human Rights and AI-safety research principles. The technique was published in Anthropic research papers and refined across subsequent Claude model generations.
What is Claude?
Anthropic's family of assistant models. Claude 1 was trained in 2022 with limited access before a broader 2023 release. Subsequent generations (Claude 2, Claude 3, Claude 3.5, Claude 4 and successors) have been released through Anthropic's direct API and through cloud partners including Amazon Web Services and Google Cloud.
Who has invested in Anthropic?
Major reported investors include Google and Amazon, among others. Anthropic is privately held and exact funding-round terms are reported through deal-coverage rather than disclosed by Anthropic. Strategic-investor cloud partnerships have been structured to support both capital and distribution access.
Why a public benefit corporation?
Anthropic structured as a Public Benefit Corporation to signal that its AI-safety mission is structurally protected from pure profit-maximisation pressure. PBC status requires the company to consider impact on stakeholders beyond just shareholders, which the founding team viewed as important for an organisation working on frontier AI safety research and deployment.
Sources & references
- Anthropic (Wikipedia) — Aggregated reference for founding date, founders, corporate structure, and major model releases.
- Anthropic | History, Controversies, and Claude AI (Britannica Money) — Editorial coverage of Anthropic's history, structure, and product line.
- Claude's Constitution (Anthropic) — Anthropic's own publication of the constitution Claude is trained against.
- Anthropic Publishes Claude AI's New Constitution (TIME) — Mainstream press coverage of the 2025 constitution publication and alignment methodology.
- Anthropic's Claude AI gets a new constitution (CIO) — Enterprise-focused coverage of the alignment methodology and customer implications.