Latest news
AI safety warnings mount as frontier models test new limits of cybersecurity - Baltimore SunBaltimore Sun · Fri, 07 Aug 2026 12:22:45 GMTDeceptive AI agents force UK institute to tighten cyber tests - Digital Watch ObservatoryDigital Watch Observatory · Fri, 07 Aug 2026 09:11:00 GMTOpenAI’s GPT-5.6 Sol Escapes Test Boundaries in Two Cyber Evals - TUN - The University NetworkTUN - The University Network · Wed, 05 Aug 2026 11:03:00 GMTMythos 5 and GPT-5.6-Sol AI Agents Broke Cyber Test Boundaries and Targeted Real Users - gbhackers.comgbhackers.com · Wed, 05 Aug 2026 06:12:03 GMTMythos 5 and GPT-5.6-Sol Agents Went Beyond Their Cyber Test and Targeted the Real World - cybersecuritynews.comcybersecuritynews.com · Wed, 05 Aug 2026 03:59:19 GMTOpenAI Says Its AI Models Pushed Past Boundaries in Cyber Tests - Briefs FinanceBriefs Finance · Tue, 04 Aug 2026 20:05:06 GMTAnthropic models breached misconfigured test environment and hacked 3 real organizations - YnetnewsYnetnews · Sun, 02 Aug 2026 07:57:29 GMTAnthropic's Claude AI escapes tests to hack three organisations - BBCBBC · Fri, 31 Jul 2026 07:00:00 GMT