Anthropic and Claude: A Full Look at the Company Betting Everything on AI Safety

Anthropic is a rare case in tech: a company whose stated mission is literally the reason it exists. While competitors race on speed and scale, Anthropic has publicly staked its identity on a different bet since day one — that powerful AI needs to be developed carefully. Here's what that actually means in practice, not just in press releases.

The history: a split from OpenAI

Anthropic was founded in 2021 by Dario and Daniela Amodei, along with a group of former OpenAI researchers. The reason for leaving: disagreement over how seriously OpenAI was taking the risks of powerful AI models. The entire company grew out of one core position — that AI safety requires a fundamentally different approach than the one they saw at their former employer. That position, not some abstract concern about the future, sits behind nearly every product decision the company has made since.

The Claude model lineup

Today's Claude family spans several models built for different jobs. Opus is the most powerful and most expensive, meant for complex tasks that need deep reasoning. Sonnet is the mid-tier model — a balance of quality and speed, and the default across most subscriptions. Haiku is the fastest, cheapest model for simple, high-volume tasks. Separately from that core naming scheme sit newer, less predictably named models — Fable and Mythos — which ship in a more limited way and often end up at the center of industry discussions, whether over export restrictions or restricted early access for select partners.

The safety philosophy: not just words

The company's core governing document is its Responsible Scaling Policy. It's a voluntary framework of AI Safety Levels (ASL-1 through ASL-4+), modeled on the biosafety level (BSL) system used by government labs. The higher a model's potential risk, the more evaluation and red-teaming it needs before launch. One detail that genuinely sets Anthropic apart from many competitors: internal safety staff hold the formal authority to halt or delay a model's release if it doesn't clear the required risk checks — meaning safety can formally override a product launch timeline.

The second core piece is Constitutional AI: instead of training a model purely on human-rated feedback (expensive and hard to scale), Anthropic trains Claude to follow an explicitly written set of principles — a "constitution" drawn from human rights documents and safety norms. The model critiques and revises its own draft answers against those principles.

Products and ecosystem

Claude isn't just the Claude.ai chat interface. Separately, there's Claude Code for writing and editing code, a developer API with its own token-based pricing, Team and Enterprise plans, Microsoft 365 integration, and Projects/Research modes for longer, more structured work on a single task.

How seriously Claude's real-world capabilities are taken shows up in Project Glasswing, Anthropic's initiative to find vulnerabilities in critical software using a restricted model, Claude Mythos Preview. That's exactly how research group Calif.io discovered "Squidbleed" (CVE-2026-47729) — a memory leak in the Squid proxy server that had hidden in the code for 29 years, surviving three decades of releases and audits, The Register reports. Anthropic has committed up to $100 million in usage credits for Mythos Preview toward efforts like this, plus $4 million in direct donations to open-source security organizations.

Pricing, and how token billing just changed

Claude's core subscription structure: a free tier, Pro at $20/month, Max at $100/month or $200/month depending on the usage limit, plus Team and Enterprise, according to the official pricing page. There's also a fresh detail that matters specifically for power users: as of July 8, 2026, the Fable 5 model stopped counting against subscription limits and moved to separate credit-based billing — $10 per million input tokens and $50 per million output tokens. Every other model in the lineup — Opus, Sonnet, Haiku — still runs within normal plan limits.

The money and the valuation

On May 28, 2026, Anthropic closed a $65 billion funding round at a $965 billion valuation — making it the most valuable private AI company in the world, ahead of OpenAI's own reported valuation, TechCrunch reports. The round was led by Altimeter Capital, Dragoneer, Greenoaks, Sequoia Capital, Capital Group, Coatue, and D1 Capital Partners. On June 1, 2026 — the same day the round closed — the company filed a confidential draft IPO prospectus, Fortune notes. Annual run-rate revenue passed $47 billion by May 2026, with Anthropic expecting a 130% revenue increase and its first operating profit. Figures like a $1.1–1.25 trillion valuation and a fall 2026 listing window are analyst and media projections, not numbers Anthropic itself has confirmed.

Anthropic and crypto

The one place this story genuinely crosses into crypto is the market's reaction to US export controls. When access to the Fable 5 and Mythos 5 models was suspended for some customers and then lifted 19 days later, the token of decentralized AI network Bittensor (TAO) jumped 30% in just 12 hours after the restrictions were lifted — the market read it as a sign that decentralized alternatives to centralized AI gain appeal exactly at moments like this.

Criticism and challenges

The company isn't without critics. Palantir CEO Alex Karp has publicly attacked the token-based monetization model that both OpenAI and Anthropic rely on: "Something has gone completely wrong," he said, CNBC reports. That's not the only source of friction: any company that names safety as its top priority inevitably draws closer scrutiny of every move it makes — from export restrictions on its models to debates over a near-trillion-dollar valuation while its financials remain private.

How Claude stacks up against the competition

We covered a broader comparison of Claude against ChatGPT, Gemini, Grok, and Perplexity — pricing, features, and who each one fits — separately in our big comparison of 2026's top 5 AI services. The core distinction worth repeating here: while competitors mostly race on feature breadth, Anthropic has systematically built its story around trust — and based on its $965 billion valuation, that bet is paying off financially, at least so far.

The takeaway

Anthropic has reached a near-trillion-dollar valuation and filed for an IPO, all while remaining the one major lab where a formal safety process can actually block a product launch. At the same time, it's already building infrastructure for a world where AI agents don't just answer questions but make payments — including tools that let people trade its stock before the IPO even happens. The practical takeaway for readers: if you're choosing between AI services based on how much you trust a company's approach to risk, not just price or speed, Claude remains the clearest choice on the market right now.

This article is for informational purposes only; pricing, plans, and company valuations may change — check Anthropic's official website for current details.