AI & Enterprise
Cisco tests LLMs for security incident reports; time saved but hallucination risk remains
Cisco shared results of an experiment using AI to write reports for security incident response training, concluding it saves time but carries many risks. Nate Forth (네이트 포르스) of Cisco Talos said large language models can produce serious inaccuracies, odd conclusions and inconsistent writing styles. Cisco identified four main error patterns and said task-specific prompts and clear source, style and format rules reduced hallucinations. Drafting time fell 50%, but session mixing and weak proofreading prompts remained problems.
