Show HN: An accurate AI password guesser based on personal information (github.com)

PassLLM has emerged as a groundbreaking targeted password guessing framework, reportedly achieving superior accuracy by 15% to 45% over existing models. Leveraging Personally Identifiable Information (PII) such as names, birthdays, and previous passwords, this framework fine-tunes large language models (LLMs) with over 7 billion parameters on vast datasets of leaked private records. The result is a system that can predict a user's likely passwords with impressive precision, guessing 12.5% to 31.6% of passwords within just 100 attempts when sufficient PII is available. Notably, it operates on consumer hardware, making it widely accessible. This framework's significance lies in its implications for cybersecurity, highlighting how easily AI can exploit personal data to breach security. By utilizing advanced techniques such as Low-Rank Adaptation (LoRA) for efficient model fine-tuning and statistical pattern recognition from real-world breaches, PassLLM raises critical awareness on the vulnerabilities associated with weak password practices. Designed for educational purposes, it demonstrates the need for enhanced password policies and security measures, pushing both individuals and organizations to reconsider their defenses against AI-assisted password cracking.