Anthropic

Developer of the Claude model family, including Claude Mythos Preview.

Role in System Card

• Developer and trainer of Claude Mythos Preview
• Publisher of the System Card (2026-04-07)
• Operator of the RSP 3.0 framework
• Partner coordinator for Project Glasswing (defensive cybersecurity program)

Key Practices

• Training data: Uses ClaudeBot web crawler following robots.txt standards; no password-protected or sign-in pages accessed
• Crowd workers: Partners with data work platforms for preference selection, safety evaluation, and adversarial testing; committed to fair compensation and ethical workplace standards
• External testing: Provides models to external groups including government organizations for evaluation on Cyber, Loss of Control, CBRN, and Harmful Manipulation
• Usage Policy: Applies to all models including limited-release ones
• European operations: Anthropic Ireland, Limited is the provider in the EEA

Stated Concerns

Anthropic explicitly flags alarm about the industry trajectory, noting the world is “on track to proceed rapidly to developing superhuman systems without stronger mechanisms in place for ensuring adequate safety across the industry as a whole.”