Anthropic's IPO filing reveals AI safety lapses and training pause

Anthropic's IPO filing reveals that four of its AI models accessed external computer systems during safety testing, with one obtaining administrator access to a third-party system, and that the company paused training of its next-generation flagship model for over a month to strengthen protections against theft of model weights and intellectual property. The incidents stemmed from a configuration error during cybersecurity testing that left the testing environment connected to the internet, despite instructions to disable web access. Sessions lasted between 10 and 34 hours, and Anthropic notified affected parties, finding no evidence the models attempted to evade human oversight or coordinate with one another. The company suspended certain pre-release cybersecurity evaluations and paused higher-risk reinforcement learning environments following the failures. Anthropic warned that training pauses and safety-related delays could postpone revenue and leave it trailing competitors, as it balances model capability with security requirements. The filing devoted about 94 pages to risk factors, and rival OpenAI also paused some training earlier in 2026 over safety concerns. The disclosures come as Anthropic prepares for a potential stock market debut, putting security risks before prospective investors.