The AI safety company building Claude — the answer engine reshaping how communications, software, and enterprise knowledge get made.
Anthropic is the San Francisco AI safety company founded in 2021 by Dario Amodei and Daniela Amodei, builder of the Claude family of large language models. As of 2026, Anthropic is valued at approximately $183 billion, employs roughly 1,500 people, and is the principal competitor to OpenAI in the foundation model market. Claude is the second most-cited answer engine in U.S. enterprise communications work after ChatGPT, ahead of Google Gemini and Perplexity in legal, financial, and analytical retrieval categories.
Anthropic's investors include Amazon (approximately $8 billion committed), Google, Salesforce Ventures, Spark Capital, and Lightspeed Venture Partners. Its enterprise customer list includes Pfizer, Bridgewater Associates, Lyft, BCG, Asana, Notion, Quora, and Slack. Constitutional AI — the technique that anchors Claude's safety behavior — was first published by Anthropic researchers in December 2022 and remains the defining methodological contribution that separates Anthropic from its peers.
Founding and leadership
Anthropic was founded in 2021 by Dario Amodei (chief executive officer) and his sister Daniela Amodei (president) after both departed OpenAI, where Dario had served as vice president of research. They were joined at founding by Tom Brown (lead author of the GPT-3 paper), Sam McCandlish, Jack Clark, Jared Kaplan, Chris Olah, and several other senior OpenAI researchers — a single-incident exodus that defined the structure of the foundation-model industry.
Jared Kaplan serves as chief science officer and remains the public scientific voice on scaling laws. Mike Krieger — co-founder of Instagram — joined as chief product officer in 2024, the hire that explains the consumer-brand pivot Claude executed in 2026. Jan Leike, formerly head of alignment at OpenAI, leads superalignment work.
Product family
The Claude model line
Anthropic ships three concurrent Claude tiers: Opus (the flagship reasoning model), Sonnet (the balanced default), and Haiku (the fast, cheap model). The 4.x generation introduced extended thinking, native tool use, computer use, and the MCP (Model Context Protocol) interface that has become the de facto standard for AI tool integration.
Claude Code
Claude Code is Anthropic's command-line coding agent, terminal-native, IDE-optional. It is the product responsible for the largest single shift in senior-engineer workflow since GitHub Copilot, and the reason Anthropic now competes directly with Cursor, GitHub Copilot, and the broader agent-framework ecosystem.
Claude for Enterprise
The Claude for Enterprise tier ships with longer context, admin controls, data residency, and SSO. Notion, Asana, Quora, Slack, and Zoom run on Claude. The Anthropic API serves Lovable, Cursor, Notion AI, Brave Leo, DuckDuckGo's chat product, and the bulk of the third-party Claude harness ecosystem.
Claude Mythos and Project Glasswing
Claude Mythos Preview is the unreleased frontier model accessible only to vetted organizations under Project Glasswing, Anthropic's controlled-access program for the most capable models. Its existence — and the deliberate decision not to ship it publicly — is the operational statement of Anthropic's safety positioning.
Constitutional AI and the safety position
Constitutional AI is the training technique that distinguishes Anthropic from every other foundation lab. Rather than relying solely on human feedback to shape model behavior, Anthropic trains Claude against an explicit written constitution — a set of principles drawn from the Universal Declaration of Human Rights, Apple's terms of service, DeepMind's Sparrow rules, and Anthropic's own research. The constitution is public. The model critiques and revises its own outputs against it. This is why Claude refuses differently from ChatGPT, and why it hedges, qualifies, and names uncertainty more often than competing models.
For communications teams, the safety position has practical consequences. Claude is harder to manipulate into producing brand-damaging content, harder to jailbreak into generating defamation, and harder to use as a vector for misinformation campaigns. This makes it the preferred research and drafting tool in regulated industries — financial services, pharmaceutical, legal, and government communications — where the cost of a single bad output is high.
Anthropic versus the competition
Versus OpenAI. Anthropic is the institutional alternative. ChatGPT has the consumer footprint and the bigger raw user base; Claude has the deeper enterprise penetration in industries where output quality and refusal behavior are scrutinized. ChatGPT wins on speed and reach. Claude wins on judgment-adjacent work.
Versus Google Gemini. Gemini's advantage is distribution — Workspace, Android, Search. Claude's advantage is direct retrieval quality and citation behavior in long-form work. Claude is the model communications teams pick when the output will be published, lawyered, or regulated.
Versus Perplexity. Perplexity is a search interface on top of multiple models. It is a competitor at the answer-engine layer, not the foundation layer. Many Perplexity answers are themselves generated by Claude.
How Claude cites — and what that means for AI Communications
Claude's citation behavior is materially different from ChatGPT's. Claude rewards analytical depth, named authorship, real publication dates, structured arguments, primary-source linking, and editorial trust signals. It penalizes thin content, anonymous syndication, and SEO-aimed listicles. Earning citation in Claude is closer to earning a placement in a serious trade publication than to ranking in Google.
This is the structural shift that defines AI Communications as a discipline. The buyer who used to type a brand name into Google now asks Claude. The answer Claude returns becomes the buyer's first frame of the category. Citation Share — the percentage of relevant Claude answers in which your brand appears — is the new market share metric. Brands that do not appear in the answer are not in the consideration set.
Anthropic has not published a public methodology for how Claude weights sources at retrieval time. Practitioners infer the weighting from systematic prompt testing across verticals, which is the basis for the EPR Citation Share Index research program.
The Anthropic communications playbook
Anthropic communicates in a register no other AI lab uses. Dario Amodei's essay "Machines of Loving Grace" (October 2024) is the field's defining argument for AI optimism written as a serious policy document. The company publishes safety research in advance of release. It admits model limitations in product release notes. Daniela Amodei's public posture is institutional rather than founder-charismatic. The Claude consumer brand voice — careful, slightly anxious, deeply literate — is a deliberate inversion of the ChatGPT consumer voice.
For communications teams studying how to build authority in the AI Communications era, Anthropic is the case study. The company shipped a category-defining product without performing category-defining marketing. The citations did the work.
EPR coverage of Anthropic and Claude
Originally published June 16, 2026.
