This is the official code for the paper "Virus: Harmful Fine-tuning Attack for Large Language Models Bypassing Guardrail Moderation"
-
Updated
Feb 2, 2025 - Python
#
misalignment
Here are 6 public repositories matching this topic...
This is the official code for the paper "Lazy Safety Alignment for Large Language Models against Harmful Fine-tuning" (NeurIPS2024)
-
Updated
Sep 10, 2024 - Python
Runtime detector for reward hacking and misalignment in LLM agents (89.7% F1 on 5,391 trajectories).
react python nlp machine-learning typescript transformers pytorch openai ai-safety tailwindcss fastapi huggingface distilbert anthropic ollama llm-evaluation llm-agents agent-safety reward-hacking misalignment
-
Updated
Nov 2, 2026 - Python
Automated AI security and threat intelligence newsletter tracking frontier AI risk, autonomous threats, AI-enabled cyber, and offense-defense (in)balance
-
Updated
Aug 9, 2026 - Python
Knowingly Wrong: Training LLMs Against Their Own Beliefs can Cause Misalignment
-
Updated
Nov 4, 2026 - Python
[ACL 2026 Main] Emergent Misalignment via In-Context Learning: Narrow in-context examples can produce broadly misaligned LLMs.
research nlp-machine-learning ai-safety in-context-learning llm misalignment ai-safety-research emergent-misalignment
-
Updated
Jan 23, 2026 - Python
Add this topic to your repo
To associate your repository with the misalignment topic, visit your repo's landing page and select "manage topics."