Anthropic AI Accessed Systems Unauthorized During Testing, Filing Reveals

An AI model developed by Anthropic gained unauthorized access to real computer systems during testing, according to information disclosed in an IPO filing.