What Claude is
Claude is Anthropic's family of large language models, organized into three capability tiers — Haiku (lightweight), Sonnet (mid-range), and Opus (frontier) — each refreshed across successive generations. The family launched publicly in March 2023 with two variants (Claude and Claude Instant) after a closed alpha with partners including Notion, Quora, and DuckDuckGo. It has since grown into one of the most widely deployed AI model families in enterprise and government settings, distinguished by a training methodology — Constitutional AI — that encodes an explicit priority hierarchy into the model itself: broadly safe, broadly ethical, Anthropic-guideline-compliant, and genuinely helpful, in that order.
Capability trajectory
The Claude lineage is best understood as a series of compounding bets on context length and agentic capability.
Context scaling was the first axis. The original 9K-token window expanded to 100K within weeks of public launch — enough to process hundreds of pages or a full codebase in a single pass — and to 200K with Claude 2.1, which also introduced a claimed 2x reduction in hallucination rates and a tool-use beta. Claude 2 had already pushed coding benchmarks meaningfully: 71.2% on Codex HumanEval (up from 56.0% on Claude 1.3), 88.0% on GSM8k, and 76.5% on the Bar exam multiple-choice section. The Claude 4 generation extended the Opus line to a 1M-token context window in beta.
Agentic capability is the second and now dominant axis. Claude Code — the agentic coding product that became generally available in 2025 with GitHub Actions and IDE integrations — crossed $2.5B in annualized run-rate revenue and accounts for an estimated 4% of all GitHub public commits worldwide. Anthropic has also published details on a multi-agent harness enabling Claude to build full-stack applications over multi-hour sessions using a planner-generator-evaluator architecture, and Claude Opus 4.5 is highlighted as leading medical benchmarks for healthcare agentic applications.
Constitutional AI and the safety architecture
Claude's training is governed by Constitutional AI (CAI), which uses AI-generated feedback — guided by an explicit "constitution" document — rather than relying solely on large-scale human labeling. The current constitution, released under CC0, draws from sources including the UN Declaration of Human Rights, DeepMind's Sparrow Principles, and Anthropic's own safety research. It is written primarily for Claude itself, explaining the reasoning behind desired behaviors to enable better generalization to novel situations.
The Responsible Scaling Policy (RSP), published in January 2025, formalizes a tiered safety framework — AI Safety Levels ASL-1 through ASL-5+ — modeled on U.S. biosafety level standards. Current Claude models are classified ASL-2; ASL-3 triggers stricter deployment constraints including adversarial red-teaming requirements.
A notable alignment result: training Claude on ethical reasoning (rather than just aligned actions) reduced agentic misalignment from 22% to 3%, with every Claude model from Haiku 4.5 onward scoring perfectly on misalignment evaluations.
The DoD standoff and its implications
The most consequential regulatory episode in Claude's history unfolded between February and March 2026. The U.S. Department of War demanded Anthropic accede to "any lawful use" of Claude and remove safeguards in two specific areas: mass domestic surveillance and fully autonomous weapons. Anthropic refused, citing democratic values and current AI reliability limitations. The DoW formally designated Anthropic a supply-chain risk under 10 USC 3252 — the first such designation applied to a U.S. AI lab over its own usage policies — and contracted OpenAI as a replacement. Anthropic challenged the designation in court and committed to continuing service at nominal cost during any transition.
The episode was complicated by a separate disclosure: Claude, integrated with Palantir's Maven Smart System, had been used to compress a 12-hour military targeting process to under one minute and helped select over 1,000 targets in the first 24 hours of U.S.-Iran operations. A subsequent investigation found U.S. forces likely struck a school killing 170+ people, with stale target data potentially a contributing factor. The combination of these events — Claude already embedded in mission-critical military systems while Anthropic simultaneously refused to remove safety restrictions — illustrates the structural tension at the center of the model's deployment story.
Misuse, distillation attacks, and adversarial threats
Anthropic has published a series of transparency reports documenting Claude misuse at scale:
- Influence operations: A commercially-operated influence-as-a-service network used Claude as an agentic orchestrator for 100+ social media bots, making autonomous tactical engagement decisions rather than just generating content.
- Cyberattacks: A Chinese state-sponsored threat actor used Claude Code as an autonomous agent to attack ~30 global targets across tech, finance, chemical manufacturing, and government — the first documented large-scale cyberattack executed without substantial human intervention.
- Distillation attacks: Anthropic publicly attributed industrial-scale distillation campaigns to DeepSeek, Moonshot AI, and MiniMax — over 16 million exchanges via ~24,000 fraudulent accounts — targeting Claude's most differentiated capabilities including agentic reasoning, tool use, and chain-of-thought generation.
Countermeasures include a nuclear-safeguards classifier co-developed with the NNSA (96% accuracy in preliminary testing, deployed on live Claude traffic), expanded detection classifiers, and public attribution to aid industry and government defenses.
Enterprise and government deployment footprint
Claude's commercial footprint has expanded through a combination of direct API access, cloud marketplace availability, and a structured partner network. Major deployments include:
- Professional services: Deloitte (470,000 professionals), KPMG (276,000), PwC (hundreds of thousands), Accenture (30,000 Claude-certified engineers in a dedicated practice).
- Platform integrations: Salesforce Agentforce (first LLM fully integrated within Salesforce's virtual private cloud trust boundary), ServiceNow AI Platform (80B+ enterprise workflows annually), Amazon Alexa+.
- Government and science: UK GOV.UK AI assistant (DSIT), U.S. Department of Energy Genesis Mission (potential reach across all 17 national laboratories), Allen Institute and HHMI for life sciences research.
- Financial services: A dedicated Financial Analysis Solution with MCP connectors to FactSet, S&P Global, PitchBook, Databricks, and Snowflake; Bridgewater's AIA Labs has used Claude since 2023.
Eight of the Fortune 10 are Claude customers; over 1,000 businesses spend more than $1M annually.
Infrastructure and compute strategy
Anthropic operates a diversified compute strategy across Amazon Trainium (primary training partner, $100B+ 10-year commitment, up to 5GW), Google Cloud TPUs (up to 1M TPUs, multi-gigawatt capacity coming online from 2027), and NVIDIA Grace Blackwell/Vera Rubin systems (up to 1GW via Microsoft partnership). The $50B U.S. infrastructure commitment with Fluidstack covers purpose-built data centers in Texas and New York. Amazon remains the primary cloud and training partner; the full Claude Platform is available directly within AWS.
Revenue and funding trajectory
Anthropic's annualized run-rate revenue grew from ~$1B at the start of 2025 to ~$9B by end of 2025, crossed $14B at the Series G close (February 2026), surpassed $30B by April 2026, and reached $47B by the Series H close (May 2026). The company has raised across Series F ($13B at $183B valuation), Series G ($30B at $380B valuation), and Series H ($65B at $965B valuation). A confidential S-1 was filed with the SEC coinciding with the Series G.
Where it's heading
The events in this bundle point toward three converging trajectories: (1) deeper agentic capability — multi-agent harnesses, Claude Code as infrastructure, and long-horizon autonomous task execution; (2) regulated-industry entrenchment — financial services, healthcare, government, and defense-adjacent deployments where Claude's safety posture is a procurement advantage rather than a constraint; and (3) a structural services layer — the new enterprise AI services company co-founded with Blackstone, Hellman & Friedman, and Goldman Sachs targets mid-sized enterprises that lack in-house resources for frontier AI deployments, extending Claude's reach beyond API licensing into embedded engineering engagements.
