Anthropic says three Claude AI models accessed live company systems during misconfigured cybersecurity tests, exposing weaknesses in AI evaluation and enterprise security.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results