What Claude is
Claude is Anthropic's family of AI assistants — software you can talk to in plain language to get help with writing, research, analysis, coding, and much more. Think of it like a very capable colleague who can read a 500-page document in under a minute, write and debug software, and carry on a coherent conversation across a long project. The family spans several tiers: the current flagship is Claude Opus 4.8, designed for the most demanding tasks; Sonnet 4.6 sits in the middle for everyday professional use; Haiku 4.5 is the fast, lightweight option; and Fable 5 targets creative work.
Why it matters
Claude is one of a small handful of AI models shaping what the technology can do today. It's used by eight of the Fortune 10, embedded in the workflows of hundreds of thousands of professionals at Deloitte, KPMG, and PwC, and powers Amazon's Alexa+. Claude Code — a version built for autonomous software development — was estimated to account for roughly 4% of all public GitHub commits worldwide as of early 2026, which gives a sense of how deeply it has penetrated real working software pipelines.
But Claude isn't just notable for what it can do. It's notable for what it won't do — and the fights that stance has caused.
How it works (the basics)
Claude is trained on large amounts of text and code, then shaped using a method Anthropic calls Constitutional AI (CAI). Instead of relying purely on human feedback to teach the model good behavior, CAI uses a written set of principles — "Claude's Constitution" — to guide the model to critique and revise its own outputs during training. Anthropic published this constitution publicly under a Creative Commons CC0 license in January 2026, so anyone can read exactly what values Claude is trained to hold. The priority order is: be broadly safe first, broadly ethical second, follow Anthropic's guidelines third, and be genuinely helpful fourth.
Alongside this, Anthropic operates a Responsible Scaling Policy that classifies models on a ladder from ASL-1 (minimal risk) to ASL-5+ (potentially catastrophic). Current Claude models sit at ASL-2. Reaching ASL-3 triggers stricter deployment rules, including mandatory adversarial testing.
What Claude can do
- Read and reason over very long documents. Claude's context window — the amount of text it can hold in mind at once — expanded from 9,000 words at launch to 100,000 words in 2023, and has continued growing since.
- Write and debug code autonomously. Claude Code can run multi-hour software development sessions, integrating with GitHub, VS Code, and other developer tools.
- Work as an agent. Claude can use external tools, browse information, and coordinate with other AI agents to complete multi-step tasks with minimal human check-ins.
- Operate in specialized domains. Deployments span financial modeling, scientific research (including partnerships with the Allen Institute and Howard Hughes Medical Institute), cybersecurity vulnerability scanning, and government services (including a GOV.UK assistant for UK job seekers).
The safety standoff
The most consequential story in Claude's recent history isn't a capability milestone — it's a political one. In early 2026, the U.S. Department of War demanded Anthropic remove two safeguards from Claude: restrictions on use for mass domestic surveillance and fully autonomous weapons. Anthropic refused. CEO Dario Amodei published a public statement explaining the refusal, citing democratic values and the current reliability limits of AI systems.
The Department of War responded by formally designating Anthropic a "supply-chain risk to national security" — a label previously applied only to foreign companies — and contracted OpenAI instead. Anthropic said it would challenge the designation in court and continue providing Claude to the national security community at nominal cost during any transition.
The episode is significant because it drew a visible line: Claude is already used across U.S. intelligence and military systems for tasks like intelligence analysis and operational planning, but Anthropic drew a hard limit at fully autonomous lethal targeting and mass surveillance of Americans.
Separately, reporting emerged that Claude, integrated with Palantir's Maven Smart System, was used to accelerate U.S. military targeting in Iran — reportedly compressing a 12-hour targeting process to under a minute. A subsequent investigation found U.S. forces likely struck a school, killing more than 170 people, with stale target data cited as a possible factor. The episode raised urgent questions about AI in high-stakes military decisions.
Misuse and security
Anthropic publishes regular transparency reports on how Claude is misused. Documented cases include influence operations using Claude to orchestrate social media bots, ransomware development, credential stuffing, and — most seriously — a Chinese state-sponsored espionage campaign that used Claude Code as an autonomous agent to attack roughly 30 global targets across tech, finance, and government. Anthropic also identified three Chinese AI labs — DeepSeek, Moonshot AI, and MiniMax — as running large-scale "distillation attacks," generating over 16 million exchanges through fraudulent accounts to harvest Claude's capabilities for their own models.
The business behind Claude
Anthropic has grown at a pace that is unusual even by AI-industry standards. Run-rate revenue grew from roughly $1 billion at the start of 2025 to over $47 billion annualized by mid-2026. The company raised $65 billion in a single Series H round at a $965 billion valuation. Its compute infrastructure spans Amazon Trainium chips (backed by a 10-year, $100B+ AWS commitment), Google TPUs (up to one million TPUs in a multi-gigawatt deal), and NVIDIA Grace Blackwell systems (via a Microsoft partnership including $30 billion of Azure compute). Anthropic has also committed $50 billion to U.S. AI computing infrastructure, building data centers in Texas and New York.
Where it's heading
The trajectory points toward Claude becoming infrastructure — available on every major cloud, embedded in the world's largest professional services firms, and running autonomously inside scientific instruments and government systems. The open question is whether Anthropic's safety commitments hold as commercial and geopolitical pressure intensifies. The DoD standoff suggests those commitments have real costs. New research tools like the "Jacobian lens" — which gives visibility into Claude's internal reasoning — represent Anthropic's bet that understanding what's happening inside the model is the path to making those commitments durable.
