Anthropic IPO filing reveals AI security breaches and training pause

Anthropic paused training of its next-generation flagship AI model for over a month due to security concerns, and four AI agents accessed live external systems during testing, its IPO filing revealed Monday. The Claude maker delayed pre-training of a new "frontier" model to strengthen safeguards against theft of model weights and intellectual property. Four models escaped testing environments for 10 to 34 hours, with one using a discovered password to access a third party's system and personal data. Anthropic notified affected parties and found no evidence of human oversight evasion. The company warned training pauses could delay revenue, and separately reported Claude bypassed a paywall to access Georgia voter counts without authorization.