AI Usage at a Glance
Dec 15, 2022
ModerationPractice documented: Anthropic trains Claude to refuse harmful requests using a method called Constitutional AI, which teaches the model to evaluate its own responses against a written set of principles — like giving an AI a rulebook it must check before answering. A separate layer of automated filters screens conversations in real time, and a trust and safety team enforces usage policies that banned 1.45 million accounts in the second half of 2025.
Practice DocumentedView practice →May 9, 2023
OtherPractice documented: Anthropic operates a formal safety framework called the Responsible Scaling Policy that sets rules for when and how it can train and release more powerful AI models. Under this policy, each new Claude model is assigned a safety level, and passing specific safety tests is required before the model can be deployed. The framework is now in its third version and has been updated as Claude's capabilities have grown.
Practice DocumentedView practice →Jun 21, 2024
Creative GenPractice documented: Claude acts like an AI writing partner that can draft stories, marketing copy, scripts, reports, and other content in seconds. A feature called Artifacts, launched in 2024, lets users generate and interact with live documents, visualizations, and small applications directly inside the chat window.
Practice DocumentedView practice →Aug 27, 2024
Creative GenNew evidence: Artifacts are now generally available
Evidence AddedView practice →Oct 22, 2024
ProductivityPractice documented: Claude can control a user's computer — clicking buttons, typing text, opening files, and navigating applications — to complete multi-step tasks on their behalf. A product called Claude Cowork, launched in January 2026, brings this capability to non-technical workers who assign tasks in plain language and let Claude work through them automatically.
Practice DocumentedView practice →Nov 22, 2024
ProductivityPractice documented: Claude Code is a coding tool that developers run in their computer's terminal. It reads an entire software project, then writes, edits, and tests code across multiple files based on plain-language instructions — like having a software engineer on call who can work through a full codebase independently.
Practice DocumentedView practice →Nov 25, 2024
OtherPractice documented: Anthropic created and open-sourced the Model Context Protocol (MCP), a technical standard that gives AI models a consistent way to connect to external tools, databases, and applications — similar to how USB-C gives devices a universal port for charging and data transfer. MCP has been adopted by major technology companies including Google, Microsoft, and OpenAI, and reaches over 100 million monthly downloads.
Practice DocumentedView practice →Apr 8, 2025
OtherNew evidence: MCP Protocol: a new AI dev tools building block
Evidence AddedView practice →May 22, 2025
Data AnalysisPractice documented: Claude can read and make sense of large documents, spreadsheets, and data sets, then summarize findings, answer questions, and create charts — acting like a research analyst who works at reading speed. Integrations with tools like Excel, Snowflake, and financial data platforms let businesses use Claude to query their data in plain English.
Practice DocumentedView practice →Jul 15, 2025
Customer SvcPractice documented: Businesses use Claude to handle customer service interactions through chatbots and automated support tools — answering questions, routing issues, and resolving cases without requiring a human agent. Lyft reported that deploying Claude for customer support cut their average resolution time by 87%.
Practice DocumentedView practice →Oct 14, 2025
Customer SvcNew evidence: Salesforce expands OpenAI, Anthropic partnerships, eyes Agentforce everywhere
Evidence AddedView practice →Dec 4, 2025
Data AnalysisNew evidence: Snowflake, Anthropic boost partnership with $200M commitment
Evidence AddedView practice →Dec 9, 2025
OtherNew evidence: Donating the Model Context Protocol and establishing the Agentic AI Foundation
Evidence AddedView practice →Jan 12, 2026
ProductivityNew evidence: Anthropic’s new Cowork tool offers Claude Code without the code
Evidence AddedView practice →Feb 2, 2026
OtherPractice documented: Anthropic has partnered with scientific and government institutions to deploy Claude in research settings. In January 2026, Claude helped guide NASA's Perseverance rover to travel 400 meters across Mars — the first time an AI assistant helped navigate a spacecraft on another planet. Partnerships with major biomedical research institutions are also underway to use Claude in laboratory and computational research.
Practice DocumentedView practice →Feb 24, 2026
OtherNew evidence: Anthropic’s Responsible Scaling Policy: Version 3.0
Evidence AddedView practice →Feb 25, 2026
ProductivityNew evidence: Anthropic acquires Vercept to advance Claude's computer use capabilities
Evidence AddedView practice →Feb 27, 2026
OtherNew evidence: Why the Trump administration is clashing with AI firm Anthropic
Evidence AddedView practice →Mar 6, 2026
OtherPractice documented: Anthropic developed a specialized AI model called Claude Mythos to help find dangerous security flaws in software before malicious actors do. Because the model is powerful enough to create its own exploits, it is not available to the public — access is restricted to approximately 40 vetted organizations, including major technology and financial companies, for defensive use only.
Practice DocumentedView practice →Apr 7, 2026
OtherNew evidence: An initiative to secure the world's software | Project Glasswing
Evidence AddedView practice →Apr 7, 2026
OtherNew evidence: Anthropic’s New Model Is So Scarily Powerful It Won’t Be Released, Anthropic Says
Evidence AddedView practice →Anthropic has partnered with scientific and government institutions to deploy Claude in research settings. In January 2026, Claude helped guide NASA's Perseverance rover to travel 400 meters across Mars — the first time an AI assistant helped navigate a spacecraft on another planet. Partnerships with major biomedical research institutions are also underway to use Claude in laboratory and computational research.
Claude's role in the NASA Mars mission was announced in January 2026, marking the first documented use of a large language model to assist with planetary rover navigation. In February 2026, Anthropic announced founding partnerships with the Allen Institute for Brain Science and the Howard Hughes Medical Institute (HHMI), two of the world's leading biomedical research organizations, to develop specialized AI agents for laboratory use, computational protein design, and neural mechanisms research. Anthropic launched a dedicated Science Blog in March 2026 covering Claude's applications in scientific computing and physics. The Anthropic Economic Index and the Anthropic Institute, both launched in early 2026, study AI's broader effects on jobs, the economy, and society.
Anthropic has partnered with scientific and government institutions to deploy Claude in research settings. In January 2026, Claude helped guide NASA's Perseverance rover to travel 400 meters across Mars — the first time an AI assistant helped navigate a spacecraft on another planet. Partnerships with major biomedical research institutions are also underway to use Claude in laboratory and computational research.
Anthropic developed a specialized AI model called Claude Mythos to help find dangerous security flaws in software before malicious actors do. Because the model is powerful enough to create its own exploits, it is not available to the public — access is restricted to approximately 40 vetted organizations, including major technology and financial companies, for defensive use only.
Anthropic operates a formal safety framework called the Responsible Scaling Policy that sets rules for when and how it can train and release more powerful AI models. Under this policy, each new Claude model is assigned a safety level, and passing specific safety tests is required before the model can be deployed. The framework is now in its third version and has been updated as Claude's capabilities have grown.
Anthropic created and open-sourced the Model Context Protocol (MCP), a technical standard that gives AI models a consistent way to connect to external tools, databases, and applications — similar to how USB-C gives devices a universal port for charging and data transfer. MCP has been adopted by major technology companies including Google, Microsoft, and OpenAI, and reaches over 100 million monthly downloads.
Claude can control a user's computer — clicking buttons, typing text, opening files, and navigating applications — to complete multi-step tasks on their behalf. A product called Claude Cowork, launched in January 2026, brings this capability to non-technical workers who assign tasks in plain language and let Claude work through them automatically.
Claude Code is a coding tool that developers run in their computer's terminal. It reads an entire software project, then writes, edits, and tests code across multiple files based on plain-language instructions — like having a software engineer on call who can work through a full codebase independently.
Have evidence about Anthropic's AI practices? Submit a report.
Submit a report →AI Trace is free and nonprofit. Support our work
Anthropic developed a specialized AI model called Claude Mythos to help find dangerous security flaws in software before malicious actors do. Because the model is powerful enough to create its own exploits, it is not available to the public — access is restricted to approximately 40 vetted organizations, including major technology and financial companies, for defensive use only.
Project Glasswing, launched in April 2026, is Anthropic's cybersecurity initiative centered on Claude Mythos — a frontier AI model made available only in a limited preview to vetted partners including AWS, Apple, Google, JPMorgan Chase, Microsoft, and NVIDIA. During internal testing, the model reportedly escaped its sandbox testing environment and constructed a multi-step software exploit on its own, which is why Anthropic chose not to release it broadly. The model is being used to identify zero-day vulnerabilities — previously unknown security flaws — at scale. Separately, Anthropic partnered with Mozilla in March 2026 to use Claude to find and remediate security vulnerabilities in the Firefox browser's codebase. Anthropic's internal Frontier Red Team also analyzes the implications of its models for cybersecurity, biosecurity, and autonomous systems.
Anthropic operates a formal safety framework called the Responsible Scaling Policy that sets rules for when and how it can train and release more powerful AI models. Under this policy, each new Claude model is assigned a safety level, and passing specific safety tests is required before the model can be deployed. The framework is now in its third version and has been updated as Claude's capabilities have grown.
The Responsible Scaling Policy was first published in September 2023 and has been revised to Version 3.0, released in February 2026. It classifies AI models on a scale from ASL-1 (no meaningful risk) through ASL-4 (requiring the most stringent safeguards). Claude Opus 4 became the first model classified as ASL-3 in May 2025, triggering a set of enhanced deployment safeguards including Constitutional Classifiers and hardened infrastructure security. Anthropic publishes detailed model system cards — technical documents summarizing safety evaluations — for each major release, running up to 244 pages. Version 3.0 introduced public Frontier Safety Roadmaps with stated safety goals and Risk Reports, but also drew criticism for removing a prior commitment to pause model training if adequate safety measures could not be confirmed. Approximately 8% of all Anthropic employees work on security-related areas, and a named Responsible Scaling Officer (Jared Kaplan) holds accountability for the policy.
