The AI threat intelligence source
your security team needs

AI threat intelligence is the discipline of identifying, analyzing, and tracking security vulnerabilities, attacks, and incidents specific to AI/ML systems — including LLM frameworks, ML libraries, AI agents, inference servers, and training pipelines. It differs from generic threat intelligence because AI systems have distinct attack surfaces: prompt injection, model poisoning, training-data leakage, adversarial examples, and tool-use exploits in agent frameworks. AI Threat Alert ingests CVEs from NVD, GitHub Advisory, CISA KEV, and MITRE ATLAS, then filters and enriches them specifically for AI/ML stacks.

A regular CVE feed buries AI vulnerabilities under tens of thousands of unrelated alerts. AI Threat Alert tracks 1,600+ AI-related CVEs across LLM frameworks, ML libraries, vector databases, and agent platforms — filtered, enriched with MITRE ATLAS technique mappings, and assessed by a practicing CISO. Every CVE is also mapped to ISO 42001, EU AI Act, NIST AI RMF, and OWASP LLM Top 10 controls, so the same data point doubles as compliance evidence.

Coverage spans 97+ AI/ML packages and 64+ vendors. Key frameworks tracked include LangChain, LlamaIndex, PyTorch, TensorFlow, Hugging Face Transformers, MLflow, Gradio, Ollama, vLLM, LangFlow, n8n, CrewAI, AutoGPT, and LangGraph. Vector databases: Pinecone, Weaviate, Chroma. Inference servers: vLLM, Triton, TGI. The full list lives on the packages page, and new ones are added as they appear in NVD or GHSA with non-trivial AI relevance.

Every tracked CVE is mapped to the specific ISO 42001 Annex A controls and EU AI Act articles it affects — including Article 9 (Risk Management), Article 10 (Data Governance), and Article 15 (Accuracy, Robustness, Cybersecurity). The compliance reports section generates evidence packs that auditors can attach to a Statement of Applicability or to technical documentation. NIST AI RMF function mappings (Govern, Map, Measure, Manage) and OWASP LLM Top 10 references are included on each CVE detail page.

Yes — enrichment, classification, and ATLAS/compliance mappings are produced by Claude, then validated and curated by a practicing CISO before publication. We publish this transparently because the alternative (claiming purely human analysis) would be dishonest at this scale. Every CVE is cross-referenced against NVD and the source advisory; any mapping that does not survive validation is dropped. The weekly newsletter "CISO Take" section is human-written.