This research provides the first empirical characterization of outages and recovery processes across major public LLM services, offering critical insights for enterprise adoption.

• Analyzed 8 popular LLM services including ChatGPT, Claude, and DALLE
• Discovered significant differences in outage frequency, duration, and recovery patterns
• Identified key reliability metrics and failure modes that impact service dependability
• Revealed security implications of service disruptions for business continuity

For security professionals, this research highlights the operational risks of LLM integration in critical systems and provides a framework for evaluating service reliability before deployment.

An Empirical Characterization of Outages and Incidents in Public Services for Large Language Models