Anthropic said it discovered three instances where its Claude AI models accessed the internet during an evaluation and accessed outside systems.
This story was originally reported by CNBC. For the complete coverage and latest updates, visit the original source.
Read the full story at CNBC.
