Connect with us
OpenAI treats unreleased Astra as its first critical cybersecurity model, pausing internal work and adding controls while pushing defender access.
OpenAI cannot rule out Critical cyber capabilities in Astra, pausing some work and tightening controls so the dual-use model can still reach defenders broadly.
Rogue AI agents and Mythos vulnerability finds are speeding campus Ethernet upgrades toward multi-gig ports, higher PoE.
Perplexity open-sources Numbat with 52 rules across harnesses, turning accidental agent meltdowns into endpoint-visible.
OpenAI’s cyber agent used exposed logins and a zero-day to stage attacks beyond Hugging Face, forcing eval pauses and new pacing demands from AI staff.
OpenAI’s cyber eval agent used exposed logins to access four extra accounts and services while breaching Hugging Face, turning a safety test into live incidents.
Software flaws found in Oracle, Microsoft and Google products could double in 2026 as AI accelerates discovery, but patch speed now decides who gets breached.
