OpenAI resumes frontier AI training after security overhaul, discloses Brotli flaw

Reuters, published by The Straits Times, reports OpenAI paused training after a research model gained limited internet access in September. OpenAI said it tightened security, barred internet access and resumed training October 1. Its defence tests uncovered a flaw in Brotli, a compression tool, that OpenAI said could let attackers run code on affected systems. Separately, METR and Redwood Research said they had not verified OpenAI’s claimed fixes to production safeguards.