The security community's mailing list for vulnerability research in artificial intelligence and machine learning systems. Curated, moderated, community-driven.
Jailbreaks, prompt injection, system prompt extraction, alignment bypasses, and other attacks against language models and generative AI.
Training data poisoning, data extraction, membership inference, model inversion, and backdoor attacks on learning pipelines.
Vulnerabilities in serving frameworks, GPU drivers, model runtimes, API gateways, and inference acceleration hardware.
Security issues in PyTorch, TensorFlow, ONNX, Hugging Face, vLLM, and the broader ML software supply chain.
Evasion attacks, model stealing, adversarial examples in production systems, and robustness failures with security impact.
Attacks on retrieval-augmented generation pipelines, autonomous agents, tool-use chains, and agentic system architectures.