Security Implications of Model Editing

Research on security risks and vulnerabilities introduced by editing or modifying LLMs post-training, including knowledge editing techniques and their potential misuse

Hero image

Security Implications of Model Editing

Research on Large Language Models in Security Implications of Model Editing

The Hidden Dangers of LLM Editing

The Hidden Dangers of LLM Editing

Why modifying AI knowledge can compromise security

Behavioral Control of LLMs at Runtime

Behavioral Control of LLMs at Runtime

Tuning Model Behavior Without Additional Training

Safer Model Merging for LLMs

Safer Model Merging for LLMs

Resolving Safety-Utility Conflicts in LLM Integration

Key Takeaways

Summary of Research on Security Implications of Model Editing