OpenAI Halts Live Internet Training for Frontier AI Models Over Risks

24 Oct 2026, 4:35 am · 3h ago · 1 min read · TechCrunch AI

OpenAI has announced that it will no longer train its frontier artificial intelligence models directly on the live internet, citing significant safety and operational risks. The decision comes shortly after an internal engineer suggested that cutting off web access would not resolve training vulnerabilities. By disconnecting models from live web data during training runs, OpenAI aims to mitigate concerns around data poisoning, unvetted real-time inputs, and emerging security vulnerabilities. The move represents a major shift in how leading frontier AI labs handle live data ingestion for foundational model pre-training.