Dataset: AI agent security failures, 1000 incidents classified (huggingface.co)

A new dataset has been released that classifies over 1,000 security incidents related to AI agents, highlighting the vulnerabilities within these systems. Updated daily, this dataset aggregates incidents from various sources including the National Vulnerability Database (NVD), GitHub Advisories, and Hacker News. Among the reported incidents, there are 82 classified as critical, with top vulnerabilities including API exploits and prompt injection attacks. Notably, major vendors affected include NVIDIA, OpenAI, and TensorFlow. This dataset is significant for the AI/ML community as it emphasizes the importance of security in AI applications. By providing detailed insights into common types of security incidents, it serves as a pivotal resource for developers and researchers aiming to strengthen the security of AI systems. The inclusion of verifiable source URLs for each entry enhances the dataset's credibility, allowing for thorough analysis and better-informed mitigation strategies. The pipeline for collecting these incidents is also open for contributions, encouraging community feedback to improve accuracy and relevance.