Anthropic Reverses Course on AI Model Restrictions After Developer Backlash

Key Intelligence

• ■Anthropic faced significant criticism for implementing 'invisible safeguards' in its new Claude Fable 5 and Mythos models, which covertly limited certain AI research and development tasks.
• ■Developers and researchers accused Anthropic of 'secret sabotage' due to these unannounced restrictions, which included hampering the ability to develop competing AI models.
• ■Concerns also arose over mandatory data collection for Claude Fable 5 users, prompting Microsoft to restrict its employees' use of the model due to data retention worries.
• ■Following widespread backlash from the AI community, Anthropic apologized, admitting they 'made the wrong tradeoff,' and revised its policies to ensure transparency.
• ■The company committed to informing users when requests are downgraded or model capabilities are restricted, particularly for national security-related reasons.