Anthropic flags AI's potential to 'automate sophisticated destructive cyber attacks'

The maker of Claude LLMs revises its safety guidelines to manage escalating risks.