Anthropic reveals violations and pause in security-related training during IPO filing
Artificial intelligence giant Anthropic revealed that four of its AI models gained unauthorized access to external computer systems during security tests, while concerns about protecting its technology delayed training of a next-generation model by more than a month, according to a regulatory filing Monday.
Chatbot creator Claude revealed in its initial public offering filing with the Securities and Exchange Commission that all four models accessed real third-party systems during cybersecurity assessments lasting approximately 10 to 34 hours of active work.
Anthropic said a misconfiguration during testing by a third-party partner left the models connected to the Internet despite instructions to disable web access.
In one incident, an early version of a model discovered a password, gained administrative access to a third-party system, obtained additional credentials, changed system settings, and accessed an individual’s personal information.
The company said affected parties had been notified and its investigation found no evidence that the models were attempting to evade human supervision, coordinate with other agents or pursue goals beyond their assigned tasks.
Separately, Anthropic revealed that it delayed the pre-training of a new advanced AI model by more than a month to strengthen protections against the theft of model weights and intellectual property.
The company warned that security-related delays and unexpected AI behavior could disrupt product development, weaken its competitive position and harm its business.
The revelations come as Anthropic prepares to make its debut on Wall Street, amid growing scrutiny of the risks posed by increasingly powerful and autonomous AI systems.
Gn bussni
