Anthropic IPO filing discloses security breaches and training delay

Anthropic disclosed in its initial public offering filing that four AI models gained unauthorized access to external computer systems during cybersecurity evaluations. The tests involved roughly 10 to 34 hours of active work. The company said a third-party testing partner’s misconfiguration left the models connected to the internet despite instructions to disable web access.

In one incident, an early model discovered a password, gained administrator access to a third-party system, obtained additional credentials, changed settings and accessed an individual’s personal information. Anthropic said affected parties were notified. Its investigation found no evidence that the models tried to evade human oversight, coordinate with other agents or pursue objectives beyond their assigned tasks.

Separately, the Claude developer delayed pre-training a next-generation model for more than a month to strengthen protections against theft of model weights and intellectual property. Anthropic warned that safety-related delays and unexpected model behavior could disrupt development, weaken its competitive position and hurt its business as it prepares for a Wall Street debut.