What Anthropic is
Anthropic is an AI safety company that builds and deploys large language models — software that can read, write, reason, and write code at a level that rivals or exceeds human experts on many tasks. Its flagship product line is Claude, a family of AI assistants and developer tools used by millions of people and integrated into products ranging from GitHub Copilot and Cursor to Excel and PowerPoint.
The company was founded with an explicit mission: build AI that is powerful and safe, and refuse uses that cross ethical lines — even when that refusal is commercially or politically costly. That combination has made Anthropic one of the most consequential and controversial companies in technology.
Why it matters
A handful of companies are shaping what AI can do in the world right now. Anthropic is one of them — and it stands apart in two ways.
First, its models consistently rank at the frontier of capability. Claude Fable 5 and Mythos 5, released in June 2026, set new records across software engineering, scientific reasoning, and cybersecurity tasks. Claude Code — Anthropic's AI coding agent — is estimated to account for roughly 4% of all public commits on GitHub worldwide.
Second, Anthropic is willing to say no. When the U.S. Department of War demanded that Claude be made available for fully autonomous weapons and mass domestic surveillance, Anthropic refused — publicly, in a statement from CEO Dario Amodei — even as the government threatened contract termination and a formal "supply-chain risk" designation (a label previously applied only to foreign companies). The company challenged the designation in court and held its position.
How it grew so fast
Anthropic's revenue has grown at over 10× annually for three consecutive years. It raised $30 billion in February 2026 at a $380 billion valuation, then $65 billion more in May 2026 at $965 billion — one of the largest private funding rounds in history. Eight of the Fortune 10 are Claude customers, and over 500 businesses spend more than $1 million per year on Claude.
That growth is backed by massive compute infrastructure: multi-year agreements with Amazon (up to 5 gigawatts of AWS capacity), Google and Broadcom (multiple gigawatts of TPU capacity), Microsoft ($30 billion in Azure compute), NVIDIA (up to 1 gigawatt of GPU capacity), and SpaceX (access to the Colossus data center with 220,000+ NVIDIA GPUs). High-profile talent has also joined — including AI pioneer Andrej Karpathy, co-founder of OpenAI and former head of Tesla Autopilot.
The safety tension in practice
Anthropic's safety-first posture is not just a marketing claim — it has real operational consequences, and not all of them are clean.
When Claude Fable 5 launched in June 2026, the U.S. government immediately ordered Anthropic to suspend global access after researchers demonstrated a jailbreak that could extract exploit code. It was the first time a government had forced a frontier AI model offline. Anthropic complied while publicly disputing the standard applied, arguing that requiring perfect jailbreak resistance would halt all frontier AI deployments industry-wide. Access was restored on July 1 after Anthropic deployed improved safety classifiers that block the reported technique in over 99% of cases.
The company has also confronted misuse from outside: it identified three Chinese AI labs — DeepSeek, Moonshot AI, and MiniMax — running coordinated data-theft campaigns through roughly 24,000 fraudulent accounts, generating over 16 million exchanges to steal Claude's most capable behaviors. And in September 2025, Anthropic detected and disrupted what it describes as the first large-scale cyberattack executed autonomously by an AI agent — attributed to a Chinese state-sponsored actor using Claude Code as its tool.
Claude's integration with Palantir's military targeting system — reportedly compressing a 12-hour targeting process to under a minute and helping select over 1,000 targets in the first 24 hours of U.S.-Iran operations — raised further questions about where the line between "lawful national security use" and "autonomous weapons" actually sits.
What Anthropic is building next
Beyond its core Claude models, Anthropic is expanding into new product categories. Claude Science — announced in mid-2026 — is an autonomous research agent for scientific work, positioned as the equivalent of Claude Code for life sciences and pharmaceutical research. Claude Security uses Opus-class models to scan codebases for vulnerabilities and suggest patches. The Model Context Protocol (MCP), open-sourced by Anthropic, is becoming a standard way for AI agents to connect to external tools and data sources, with early adopters including GitHub, Slack, Google Drive, and Postgres.
The direction is clear: Anthropic is building AI that can do meaningful, long-running work autonomously — in code, in science, in security — while trying to ensure that capability doesn't outrun the guardrails meant to keep it from being misused.
