🇺🇸 USA · Anthropic (the company)

Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: The frontier AI lab behind Claude. Founded by former OpenAI executives focused on AI safety. The encyclopedia’s preferred AI vendor — Anthropic’s “Constitutional AI” approach and careful policy commitments distinguish it.

Front-matter facts

Field	Value
Vendor / company	Anthropic PBC (San Francisco, USA) — a Public Benefit Corporation
Country / origin	🇺🇸 USA
Recommended for Australian users?	✅ Yes — Anthropic products fully accessible from AUS; AWS Bedrock Sydney offers AUS data residency for Claude
Privacy summary	API: no training by default. Consumer Claude.ai (since Sep 2025): trains by default with opt-out. Enterprise: never trains.
Investors	Google ( $3 B +), A ma z o n ($ 8B+), Salesforce, Spark Capital, Lightspeed, Menlo Ventures, others
Headcount	~1,000-2,000 (growing rapidly)
Founded	January 2021
Founders	Dario Amodei (CEO), Daniela Amodei (President), and 5 other former OpenAI senior researchers
Last reviewed	2026-06-26
Official site	https://anthropic.com

What it is

Anthropic is the AI lab that makes Claude. Founded January 2021 by Dario Amodei (former VP of Research at OpenAI) and Daniela Amodei (former VP of Safety + Policy at OpenAI) along with 5 other senior OpenAI researchers who left together to start a safety-focused frontier-AI lab.

Anthropic’s distinguishing characteristics:

Constitutional AI

A novel training approach where the AI is trained to follow a written “constitution” of principles rather than purely from human feedback
Result: Claude generally behaves more thoughtfully, refuses with explanation, and engages with hard topics carefully

Responsible Scaling Policy

Anthropic was the first frontier lab to publish a detailed framework for what dangerous capabilities would trigger safety pauses
Has been updated multiple times; serves as industry reference
Other labs (OpenAI Preparedness Framework, Google’s similar) followed Anthropic’s lead

Public Benefit Corporation

Anthropic is structured as a PBC, not a normal C-corp
Legally obligated to consider public benefit alongside shareholder returns
A specific safety-oriented governance choice

Research direction

Heavy focus on interpretability — understanding what’s happening inside neural networks
Famous research on “mechanistic interpretability” (finding features inside models)
Anthropic publishes more interpretability research than any other major lab

Products

Claude (consumer chat at claude.ai)
Claude Code (CLI coding agent)
Claude API (developer access)
Computer Use (API capability for desktop / browser agents)
MCP (Model Context Protocol — open standard for agent tool integration)
Claude Skills, Plugins, Cowork, Remote Desktop / Dispatch (Claude Code ecosystem)
Enterprise / Team / Edu tiers (claude.ai)

Why the encyclopedia recommends Anthropic

The encyclopedia leans toward Anthropic as the default high-trust AI vendor for several reasons:

Safety culture is genuine. Founder Dario Amodei is one of the most-respected voices on AI safety. Compensation structures and hiring prioritise safety researchers heavily.
Constitutional AI produces noticeably more thoughtful behaviour. Claude refuses to deceive users, explains its limitations, and is generally more honest about uncertainty than competitors.
Responsible Scaling Policy is publicly committed, audited, and updated. Anthropic isn’t just talking about safety — they’ve created accountability frameworks.
PBC structure provides some legal grounding for safety-first decisions.
Interpretability research suggests they’re trying to actually understand what their models do, not just ship products.
AWS Bedrock Sydney provides AUS data residency — practical Australian access.

This isn’t an unconditional endorsement — Anthropic is a US private company subject to the same legal regime as competitors, and the September 2025 policy change to train on consumer chats by default was a real shift. But for the Western AI frontier, the encyclopedia’s house view is that Anthropic’s care + transparency + interpretability work make Claude the default high-trust choice.

How to engage with Anthropic from Australia

As a consumer: claude.ai (free + Pro + Max + Team + Enterprise tiers)
As a developer: console.anthropic.com (Claude API)
As a coder: Claude Code (terminal CLI; bundled with Claude Pro+)
Via AWS: AWS Bedrock Sydney for Claude with AUS data residency
Via Google Cloud: Vertex AI Sydney for Claude
Via Salesforce: Salesforce uses Claude in Einstein / Agentforce

Important context

Anthropic’s growth

2021-2023: Small lab; Claude 1 / 2 released
2024-2025: Claude 3 / 3.5 / 4 family — rapid quality improvement
2026: Claude 4 / 5 family (Opus 4.8, Sonnet 4.6, Haiku 4.5, Fable 5)
Revenue grew from ~US $100 M (2023) \to U S$ 5B+ (estimated 2025) → much higher 2026

The OpenAI relationship

Founders left OpenAI in 2020-2021 over concerns about safety culture + commercial direction
Anthropic and OpenAI are now the two leading frontier labs
Generally collegial in research; commercially competitive
Both ship to many of the same enterprise customers

The Google relationship

Google invested ~US$3B+ in Anthropic
Claude available via Vertex AI alongside Gemini
Not exclusive — Anthropic also partners with Amazon (which invested more)

The Amazon relationship

Amazon’s largest single investment outside its own businesses (~US$8B+)
Claude is the default frontier model on AWS Bedrock
Amazon’s strategic AI partner

The Australian connection

Mike Cannon-Brookes (Atlassian co-founder, Aussie billionaire) has been publicly supportive of Anthropic’s safety-first positioning
Several Australian AI safety researchers work at Anthropic
AWS Bedrock Sydney makes Anthropic enterprise-ready for AUS

Privacy / data handling

See per-product entries:

Recent changes (company-level)

2026: Claude 5 family rolling out; revenue growth steep
September 2025: Consumer Claude.ai began training on conversations by default (opt-out available)
2025: Major funding rounds; Amazon investment expanded
2024: Claude 3 / 3.5 series; Computer Use launched (October)
2023: Claude 2; large Google investment
2022: Claude 1 (released to select partners)
January 2021: Anthropic founded

Gotchas

Anthropic ≠ AI safety institute — they’re a commercial company that prioritises safety; not an academic / non-profit
PBC structure has limits — public benefit is a legal consideration, not an absolute constraint
Recent training-on-chats policy change (Sep 2025) showed Anthropic’s policies can evolve under commercial pressure; not all changes are pro-user
Anthropic models are accessible via multiple paths — direct API, AWS Bedrock, Google Vertex, Salesforce; pricing / terms differ slightly
For very latest features, direct claude.ai or API first; cloud-distributor channels lag by weeks-to-months

Tech & AI, Explained

Explorer

anthropic