For Anthropic, from 🤫 hussh
Intelligence you can trust, on a layer you control.
Anthropic builds Claude, treats safety as product, and gave the world MCP. 🤫 Private Agent One speaks MCP every day. You bring the frontier intelligence. We bring the private layer that delivers it on the hardware a person owns, with their consent and a receipt for every call. We'd love to build this with you.
What Anthropic and 🤫 make possible.
You created MCP and gave it to a neutral foundation, so it could become the open rail for agents.
🤫 speaks MCP. Agent One's tools and data are MCP servers that Claude can reach, and Agent One reaches out the same way.
The person's Agent One, the expert's 🤫 Agent and Claude, all on the same consented protocol.
You lead on safety, interpretability and candor, with Constitutional AI and a public Responsible Scaling Policy.
We lead on consent: receipted, human-controlled ownership of a person's data.
Intelligence you can trust, through a layer the person controls. One story.
Regulated buyers want Claude, and they need governance before data leaves the building.
Agent One is the private, on-device client layer, with consent metering and PCHP receipts.
Claude reaches insurance, finance and healthcare inside the controls those buyers need.
Claude, safety and the open rail they built.
Anthropic is an AI safety and research company structured as a Public Benefit Corporation, with a Long-Term Benefit Trust that keeps the mission first.
Its mission: the responsible development of advanced AI for the long-term benefit of humanity.
At about a $965B post-money valuation, among the most valuable private companies in the world.
Anthropic's stated run-rate as of May 2026.
Anthropic created the Model Context Protocol in November 2024 and donated it to a Linux Foundation body in December 2025.
Frontier models across Opus, Sonnet and Haiku, plus Claude Code for agentic coding.
A published Responsible Scaling Policy and AI Safety Levels, the most explicit public safety framework in the industry.
Frontier capability with safety and openness written down in public. The kind of company a private agent wants to build with.
The Claude model family
Frontier models across Opus, Sonnet and Haiku, with large context windows, trusted for reasoning, coding and enterprise agents.
MCP, the open standard
Anthropic created the Model Context Protocol and gave it to a neutral Linux Foundation body. Building on MCP is building on open ground.
Safety and alignment
A published Responsible Scaling Policy, AI Safety Levels and Constitutional AI. Safety as product, from the start.
Interpretability
Circuit tracing and attribution graphs: the deepest look anyone has taken inside a large model.
Claude Code
The agentic-coding product and the Agent SDK behind it, the harness for real agentic work.
Enterprise reach
Claude Enterprise, the Claude API, AWS Bedrock, Google Cloud and Microsoft Foundry.
Where Anthropic is going.
Every one of these priorities has room for a private agent the person owns.
Safe frontier AI
Safety and interpretability research alongside capability, with the Responsible Scaling Policy as the public commitment.
Enterprise and agentic coding
Claude Code and the Agent SDK lead the business: agents that do work, not just summarize it.
MCP stewardship
MCP as the neutral, open rail for agents across the whole industry.
Interpretability
Understanding "the biology of a large language model" as a foundation for safety.
Beneficial acceleration
Dario Amodei's thesis: powerful AI, developed safely, compressing progress in health, science and human flourishing.
Scaling Claude
More compute across more clouds to meet demand for Claude, while holding the safety line.
The Claude lineup, the Agent SDK and MCP's new home.
What Anthropic shipped most recently, and where Agent One plugs in.
Current Claude models
A refreshed lineup across Opus, Sonnet and Haiku, with large context and strong agentic and coding performance, on the API and every major cloud.
Claude Code and the Agent SDK
The agentic-coding product and the Python and TypeScript Agent SDK: subagents, sessions and native MCP. The same harness that helps build this site.
MCP at the Agentic AI Foundation
In December 2025 Anthropic donated MCP to a Linux Foundation body co-founded with Block and OpenAI. MCP is now neutral, openly governed infrastructure.
Constitutional AI and classifiers
A published constitution and next-generation Constitutional Classifiers against jailbreaks. Safety you can read.
Buy together. Build together. Sell together. Take risk together.
We already share an open standard, MCP, and the values underneath it: consent, safety and candor. Here is what we will build on that.
Build: Claude with consent
Claude is available on the API and the AWS, Google and Microsoft marketplaces.
A person's 🤫 Agent One calls Claude with their explicit consent for reasoning, coding and agentic work, with a PCHP receipt for every call.
High-intent Claude usage, while the person keeps ownership and control of their data.
Build: MCP-native connections
Anthropic created MCP and made it neutral, open infrastructure.
Agent One exposes tools and data as MCP servers that Claude clients can reach, and reaches others in turn. The bridge for the 🤫 Yellow Pages connection engine.
A seeker's agent and an expert's 🤫 Agent meet agent to agent on the shared rail, and MCP grows.
Sell: a private client layer for Claude
Insurance, finance and healthcare want Claude, with governance.
🤫 adds owner-operated execution and consent receipts around Claude, and we are building to FedRAMP High (in pursuit).
Anthropic reaches regulated markets. The buyer gets frontier intelligence inside controls they trust.
Buy: metered, consented intelligence
Anthropic sells Claude on the API and cloud marketplaces.
We buy Claude on a metered, consent-gated basis instead of building frontier models. Buy the intelligence, own the private layer.
Durable Claude demand, and a team free to perfect the layer the person controls.
Take risk: the values bet
Anthropic champions safety, interpretability and candor.
🤫 is consent-first, candid in what it says, human-controlled and owner-operated.
Together we invest in MCP-native, consent-first agent commerce, betting that trust becomes the norm.
The people building Anthropic.
Dario Amodei
Co-Founder and CEO
Builds frontier AI that could compress decades of progress, gated by safety commitments.
Daniela Amodei
Co-Founder and President
Leads operations and the teams delivering Claude, Claude Code and the enterprise business.
Chris Olah
Co-Founder, Interpretability
Leads mechanistic interpretability: circuit tracing and attribution graphs to see what happens inside the models.
Safety and consent as product, for everyone.
Anthropic treats safety as product. 🤫 treats consent the same way. Together: intelligence you can trust, through a layer you control, on hardware you own, including for the people who need trust most. 🤫 Agent One is free for life to every federal and state government employee and every past or present service member.