This research introduces an innovative approach to protect critical power infrastructure by detecting previously unknown (zero-day) attacks in digital substations using in-context learning techniques.

• Addresses the growing challenge of novel cyber attacks on power grids using IEC-61850 communication protocols
• Overcomes limitations of traditional ML methods through specialized in-context learning approaches
• Demonstrates improved detection capabilities for previously unseen attack vectors
• Enhances resilience of critical infrastructure security without requiring complete system retraining

This advancement is crucial for energy security as it provides power utilities with more robust defenses against emerging threats that traditional systems might miss, helping prevent potentially catastrophic grid disruptions from sophisticated attackers.

Detecting Zero-Day Attacks in Digital Substations via In-Context Learning