Anthropic Responsible Scaling Policy

• Authority: Private-sector companies
• Collections: Corporate policies and commitments;Editors' Picks
• Most recent activity: Enacted
• Most recent activity date: 2024-10-15

Risk subdomain coverage

Good coverage — Comprehensive governance measures or mitigations with clear procedures relating to the risk subdomain are explicitly described.

• AI system security vulnerabilities
• Cyberattacks, weapon development or use
• Governance failure
• AI possessing dangerous capabilities
• Lack of capability or robustness

Minimal coverage — Brief mention but no specific mitigations, recommendations or governance measures relating to the risk subdomain are described.

• Exposure to toxic content
• Disinformation, surveillance, and influence at scale
• Fraud, scams, and targeted manipulation
• Competitive dynamics
• AI pursuing its own goals

All other risk subdomains had no coverage in this document.

Risk Coverage Summary

The document has good coverage of approximately 8-10 subdomains, with strong focus on malicious actors (4.1, 4.2, 4.3), AI system security (2.2), competitive dynamics (6.4), governance failure (6.5), and AI safety failures (7.1, 7.2, 7.3). Coverage is concentrated in security, misuse prevention, and AI safety domains.

Sector coverage

Good coverage — Comprehensive governance measures or mitigations with clear procedures relating to the sector are explicitly described.

• Information
• Scientific Research & Development Services

All other sectors had no coverage in this document.

Sectors Summary

This is an internal corporate policy governing Anthropic's own operations as an AI development company. The primary sectors governed are Information (AI/technology development) and Scientific Research and Development Services (AI research). The policy does not regulate external sectors but rather establishes governance for the company's AI model development, deployment, and security practices.

Legislative Status

Other

Legislative Status Summary

This is an internal corporate policy document establishing Anthropic's own responsible scaling framework. It uses mandatory language ('will', 'must', 'shall') but these obligations are self-imposed by the company rather than legally binding external requirements. The document establishes internal governance structures, reporting mechanisms, and decision-making processes for the company's AI development activities.

Actors

Proposer

• Entity Types: AI_Developer
• Entities: Anthropic
• Summary: Anthropic is the author and proposer of this Responsible Scaling Policy. The document is their internal governance framework for AI development and deployment, as evidenced by references to 'our models', 'our risk mitigation strategy', and internal organizational structures.

Target

• Entity Types: AI_Developer
• Entities: Anthropic
• Summary: The policy applies to Anthropic itself - governing its own AI model development, deployment, and security practices. It regulates internal activities including model training, deployment decisions, security measures, and employee conduct.

Enforcer

• Entity Types: AI_Developer
• Entities: Responsible Scaling Officer; CEO; Board of Directors; Long-Term Benefit Trust
• Summary: Internal enforcement is conducted by designated roles within Anthropic's governance structure. The Responsible Scaling Officer has primary responsibility for ensuring policy compliance, with oversight from the CEO, Board of Directors, and Long-Term Benefit Trust.

Monitor

• Entity Types: AI_Developer, AI_Governance_Actor
• Entities: Responsible Scaling Officer; Board of Directors; Long-Term Benefit Trust; third-party reviewers; U.S. Government entity
• Summary: Monitoring involves both internal oversight (RSO, Board, LTBT) and external mechanisms (third-party reviews, government notification). The policy establishes regular assessment cycles, reporting requirements, and independent audits.

AI Lifecycle Stages

Lifecycle stages covered by the documents, as defined in the OECD Framework for the Classification of AI Systems

• Plan and Design
• Build and Use Model
• Verify and Validate
• Deploy
• Operate and Monitor

AI Lifecycle Summary

The document comprehensively covers Build and Use Model, Verify and Validate, Deploy, and Operate and Monitor stages. It focuses extensively on testing models for dangerous capabilities, implementing safeguards before deployment, and ongoing monitoring post-deployment. Plan and Design is implicitly covered through capability threshold planning, while data collection is not substantially addressed.

Technical Scope

Terms relating to technical scope that are mentioned or implicitly referenced in the document

• AI Models
• Frontier AI
• AI Systems
• Compute Threshold
• Open Weight / Open Source

Technical Scope Summary

The document explicitly covers AI models and AI systems, with detailed focus on frontier AI models. It does not explicitly mention general purpose AI, task-specific AI, foundation models, generative AI, or predictive AI by those specific terms. It extensively discusses compute thresholds using FLOP metrics and Effective Compute. Open-weight models are mentioned in the context of release decisions.