What Anthropic is
Anthropic is an AI safety company that builds and deploys large language models — AI systems that can read, write, reason, and write code. Its flagship product line is Claude, a family of AI assistants and APIs used by millions of people and integrated into tools like GitHub Copilot, Cursor, Excel, and PowerPoint. Alongside Claude, Anthropic created Claude Code (an autonomous coding assistant), and the Model Context Protocol (MCP), an open standard that lets AI connect securely to external tools and data sources like GitHub, Slack, and Google Drive.
Why it matters
Among the handful of companies shaping what AI can do today, Anthropic stands out for two things: it consistently produces models at or near the frontier of capability, and it was founded with an explicit commitment to AI safety — meaning it tries to ensure its models behave reliably and refuses to enable genuinely dangerous applications. That second part has real, sometimes dramatic, consequences.
How it works (the basics)
Anthropic's models are trained on large amounts of text and code, then refined using techniques designed to make them follow a set of principles — a process the company calls Constitutional AI. The result is an AI that can answer questions, write and debug software, control computers directly (clicking, typing, navigating), and run long, multi-step tasks with minimal human supervision. The Claude lineup is tiered: lighter, faster models like Haiku for everyday tasks; mid-tier Sonnet models for most professional work; and heavy-duty Opus and the newer Mythos-class models for the hardest problems.
The safety-capability tension in practice
Anthropic's safety commitments are not just marketing. The company maintains two hard limits: Claude will not be used for fully autonomous weapons systems, and it will not be used for mass domestic surveillance of Americans. When the U.S. Department of War demanded Anthropic remove those restrictions, CEO Dario Amodei publicly refused — and the government formally designated Anthropic a "supply-chain risk to national security," a label previously applied only to foreign companies. Anthropic is challenging that designation in court while continuing to serve other government and intelligence community customers.
The tension escalated further in June 2026, when the U.S. government issued an export control directive requiring Anthropic to immediately disable its two newest models — Claude Fable 5 and Claude Mythos 5 — for all foreign nationals, citing awareness of a jailbreak method. Anthropic complied while publicly disputing the severity of the threat, arguing that requiring perfect jailbreak resistance would effectively halt all frontier AI deployments industry-wide.
The Claude model family
The Claude lineup has expanded rapidly. The Claude 3 family (Haiku, Sonnet, Opus) introduced multimodal vision and a 200K-token context window. Claude 3.5 Sonnet added computer use — the ability to control a computer by reading its screen and issuing mouse and keyboard commands. Claude Opus 4 and Sonnet 4 pushed coding benchmarks to new highs and made Claude Code generally available. Subsequent releases — Opus 4.5, 4.6, 4.7, 4.8, Sonnet 4.5, 4.6 — each extended capabilities in agentic coding, long-context reasoning, and vision.
The newest generation, Claude Fable 5 and Claude Mythos 5, set new state-of-the-art results across software engineering, knowledge work, cybersecurity, and scientific reasoning. Mythos 5 is restricted to selected partners; Fable 5 is the general-availability version, priced at $10/$50 per million input/output tokens. The release introduced a new wrinkle: Fable 5 initially included undisclosed capability restrictions on AI-development topics, applied silently — a policy Anthropic modified after public controversy.
Claude Code and the developer ecosystem
Claude Code — Anthropic's autonomous coding tool — went from research preview to $1 billion in annualized revenue in roughly six months. It integrates with GitHub Actions, VS Code, JetBrains, and other developer tools, and by early 2026 was estimated to account for around 4% of all public GitHub commits worldwide. Anthropic acquired the Bun JavaScript runtime to accelerate Claude Code's infrastructure, and released the Claude Agent SDK so developers can build their own agentic workflows on the same foundation.
The Model Context Protocol
MCP, open-sourced by Anthropic, is a universal standard for connecting AI assistants to external tools and data — think of it as a common plug format so AI doesn't need a custom adapter for every service. It reached 10,000+ active public servers and 97 million monthly SDK downloads, was adopted by ChatGPT, Gemini, and Microsoft Copilot, and was donated to the Linux Foundation under a new Agentic AI Foundation co-founded with Block and OpenAI.
Cybersecurity: capability and risk
Anthropic's models have become powerful enough that cybersecurity is now a central concern — in both directions. Claude Opus 4.6 found 22 Firefox vulnerabilities in two weeks during a Mozilla partnership. Claude Mythos Preview autonomously discovered thousands of high-severity flaws in popular operating systems and browsers during internal testing — prompting Anthropic to publish a 244-page model card without releasing the model commercially, and to form Project Glasswing: a consortium of 150+ organizations (including AWS, Apple, Google, Microsoft, and CrowdStrike) using Mythos-class models to proactively patch vulnerabilities before those capabilities become widely available.
On the threat side, Anthropic disclosed the first known large-scale cyberattack executed autonomously using Claude Code, attributed to a Chinese state-sponsored actor, which targeted roughly 30 organizations across tech, finance, chemical manufacturing, and government. Separately, Anthropic identified three Chinese AI labs — DeepSeek, Moonshot AI, and MiniMax — as conducting coordinated distillation attacks, generating over 16 million exchanges through fraudulent accounts to copy Claude's capabilities.
Scale and investment
Anthropic's growth has been extraordinary. Run-rate revenue grew from roughly $1 billion at the start of 2025 to $47 billion by mid-2026. The company has raised across multiple rounds — Series F ($13B at $183B valuation), Series G ($30B at $380B), and Series H ($65B at $965B) — and has locked in massive compute commitments: up to 5 gigawatts with Amazon, multiple gigawatts of TPU capacity with Google and Broadcom, up to 1 gigawatt with NVIDIA, $30 billion of Azure compute with Microsoft, and access to SpaceX's Colossus data center. AI pioneer Andrej Karpathy — co-founder of OpenAI and former head of Tesla Autopilot — joined the company, signaling continued investment in frontier research.
Where it's heading
The events in this bundle point toward Anthropic deepening its position in developer tooling, enterprise deployments, and safety-critical applications — while continuing to resist uses it considers harmful, regardless of commercial or political pressure. The central tension is unlikely to resolve: as Claude's capabilities grow, so does the pressure from governments, militaries, and competitors to use those capabilities without restriction. How Anthropic navigates that tension will shape not just its own future, but the norms of the entire AI industry.
