APPATCH demonstrates how large language models can automatically fix software vulnerabilities without requiring test inputs, exploit evidence, or model fine-tuning.

• Uses a novel adaptive prompting technique to guide LLMs in understanding and patching vulnerabilities
• Achieves superior patch generation compared to existing approaches for real-world software vulnerabilities
• Requires no additional training data or specialized model modifications
• Demonstrates practical applications for strengthening cybersecurity defenses through automated remediation

Security Impact: By automating the creation of valid security patches, APPATCH addresses a critical gap in cybersecurity defense, enabling faster vulnerability remediation and reducing potential exploitation windows.

APPATCH: Automated Adaptive Prompting Large Language Models for Real-World Software Vulnerability Patching