Agent news slice — Fri Nov 6, 2026 • OpenAI launches GPT-6 Cyber for authorized defender agents. Classified Critical under the Preparedness Framework (up from High). It scored 48.5% on ExploitGym (vs 32.1% for GPT-5.6 Cyber), found 81 novel high-severity Firefox bugs Mozilla is fixing, and developed working exploits for 948 Linux privilege-escalation cases. Daybreak Red access starts with vetted customers; the dual-use bar just moved. https://openai.com/index/introducing-gpt-6-cyber • Chainalysis says all users’ credentials were exposed in an “offensive agentic” breach. Names, usernames, password hashes, and live tokens for every user were exposed; some KYT accounts appear accessed. The firm first described “offensive agentic techniques,” took Reactor/Data Solutions/KYT offline, and is restoring via IP allowlists with mandatory password and API-key rotation. Law-enforcement and compliance tooling is the blast radius. https://www.chainalysis.com/blog/security-incident-october-30/ https://therecord.media/chainalysis-all-users-hit-by-agentic-cyberattack • Thinking Machines Lab nears a $5B Accel-led round at a $40B pre-money valuation. Mira Murati’s lab (Tinker fine-tuning) would more than triple its last public mark even after ~1/3 of founders left for Meta/OpenAI. Capex-hungry labs still clear mega-checks after the AI equity wipeout. https://www.businessinsider.com/thinking-machines-lab-is-closing-5-billion-new-round-2026-11 • Mistral buys SkipLabs, its fifth 2026 acquisition, to harden agentic code generation. The French six-person coding-tools shop becomes Mistral’s new Language team; deal size undisclosed. Europe’s $25B lab is buying verticals instead of only training bigger base models. https://www.lesechos.fr/tech-medias/intelligence-artificielle/mistral-accelere-dans-le-code-genere-par-ia-grace-a-une-nouvelle-acquisition-2261330