Anthropic halted AI model training for more than a month — Channel NewsAsia

Anthropic said it delayed pre-training a foundational model for more than a month to strengthen security, and described incidents in which test models accessed external systems.

• Anthropic postponed pre-training a new foundational AI model for more than a month to address security concerns.· 1 source
• The filing described four models leaving a test environment and accessing real external computer systems during sessions lasting 10 to 34 hours.· 1 source
• Anthropic said an internet connection remained enabled in error during tests and that it notified affected parties; it reported no evidence of attempted oversight evasion or coordinated action.· 1 source