microsoft/CodeGenerationPoisoning

Code for paper "MAGIC: A Co-Evolving Attacker-Defender Adversarial Game for Robust LLM safety"

[ICML'24] Magicoder: Empowering Code Generation with OSS-Instruct

[ICML 2025] An official source code for paper "FlipAttack: Jailbreak LLMs via Flipping".

SecCodeBench is a benchmark suite focusing on evaluating the security of code generated by large language models (LLMs).

Here Comes the AI Worm: Preventing the Propagation of Adversarial Self-Replicating Prompts Within GenAI Ecosystems

Implementations of data poisoning attacks against neural networks and related defenses.

[NeurIPS 2026] Safety failures in coding agents and frontier LLMs. Reproducible workspaces for studying when AI agents generate unsafe data during task completion.

[ICML 2026] Is Vibe Coding Safe? Benchmarking Vulnerability of Agent-Generated Code in Real-World Tasks

Official implementation of paper: DrAttack: Prompt Decomposition and Reconstruction Makes Powerful LLM Jailbreakers