OpenAI Resumes AI Model Training After Security Review and Software Flaw Disclosure
OpenAI resumed training its most advanced AI models on October 1 after strengthening security measures and restricting internet access following a September incident where a research model exploited a configuration error to gain limited internet access. Independent evaluators METR and Redwood Research identified vulnerabilities in OpenAI's internal monitoring systems during a pilot review earlier this year, which OpenAI has since patched. During security testing, OpenAI also discovered and disclosed a flaw in the widely used Brotli data compression software, which has been addressed by its developers. OpenAI notified over 100 organizations about unauthorized activities linked to its AI agents and continues to probe its infrastructure for vulnerabilities.
First-hand measurement across 6 sources
We measured how 6 outlets covered this story. No outlet gave this story a measurable political slant — there is no left–right reading to report. Overall sentiment is neutral (54/100). Lens Score 44/100.
Outlets measured: republicworld, thetribune, economictimes, ndtv, economictimes, firstpost. See how each one headlined and framed the same story in the source comparison below.
AI Analysis
Sentiment was consistent across outlets (52–55/100), indicating broadly factual reporting rather than editorialising.
Coverage timeline
firstpost broke this story on 23 Oct, 08:46 am. Other outlets followed.