Skip to content
InfoNewsLLM-specific

ThreatsDay: GPT-6.2 Jailbreak Flaws, BubbleLoader PNG Malware, $500M Ransom Demands + 17 More Stories

Published
Record updated
View JSON

Summary

Google's DeepMind division unveiled Gemini 4 Neon, which it says produces more secure code than Carbon on AutoBaxBench, reaching 76% on the hard tier. OpenAI introduced GPT-6.2 Astra and GPT-6.2 Sol, which it says meet the Critical cybersecurity threshold under its Preparedness Framework. The U.K. AI Security Institute reported a universal jailbreak against two pre-release GPT-6.2 Astra versions in half the queries needed for GPT-6 Astra, and OpenAI has since addressed the regression.