Cisco Reports AI Limitations in Security Incident Reporting

• •Cisco found AI-generated reports can contain significant inaccuracies and inconsistencies.
• •Using granular prompts and fixed sources can mitigate some AI-related issues.
• •Human oversight is necessary to ensure the quality of AI-generated security reports.

Cisco conducted a study on using AI to generate security incident reports and found significant inaccuracies and inconsistencies. The company highlighted that large language models (LLMs) can produce unusual conclusions and varying writing styles due to their probabilistic nature. Key issues included cross-contamination of content when multiple reports were generated in a single session and challenges with formatting and structure. Cisco recommends using granular prompts and fixed source documents to improve accuracy. Despite these challenges, they noted a 50% reduction in time to draft reports, although the AI-generated recommendations often lacked actionable insights. The findings suggest that human oversight is essential for ensuring report quality.