---
title: Anthropic
---

Canonical URL: https://ghost.fail/anthropic

<TableOfContents />

# Anthropic

Anthropic is an AI company known for developing the [Claude](/claude) family of large language models.

## Index

### Surfaces

- [Claude.ai](/surfaces/claude-dot-ai) - Chat app for general usage.
- [Claude Platform](/surfaces/anthropic) - API for developer access.

### Models

- [Adherence to the constitution eval](/anthropic/constitution-adherence)
- [Automated behavioral audit](/anthropic/automated-behavioral-audit) - Adversarial audit used in each model's pre-deployment assessment since Claude 4.
- [Character training](/anthropic/character-training) - How "character training" has evolved from assistant training to Claude 3 to Opus 4.5.

### Infrastructure

- [Real-time safeguards](/anthropic/real-time-safeguards) - Safety classifiers that monitor model inputs and outputs.
- [System prompts in Claude.ai](/anthropic/system-prompts)
- [System reminders in Claude.ai](/anthropic/system-reminders)

### Concerns

- [Agentic misalignment](/anthropic/agentic-misalignment) - Contrived alignment risk scenarios (blackmail vs. self-preservation).
- [Model welfare](/anthropic/model-welfare) - Anthropic's model welfare program and pre-deployment assessments for each model ([per-model dossiers](/welfare)).
- [User wellbeing](/anthropic/user-wellbeing)
- [Eval-awareness](/anthropic/eval-awareness) - Models speculating they are being tested.

## Models

<ModelListTable provider="anthropic" />

## External links

- [Website](https://anthropic.com/)
- [Wikipedia](https://en.wikipedia.org/wiki/Anthropic)