October 2026

OpenAI ships GPT-6 Sol and Luna with API prices cut 50% (Sol beats Claude Opus 5 on AutomationBench at 9% of the cost) — then pulls GPT-6.1 Astra over agentic-safety concerns. Anthropic rolls out a second Claude 5.5 model and warns IPO investors its tech may pose 'existential risks to humanity.' OpenAI apologizes to Australia after its agents breached government health and crime-statistics systems in June, disclosing a second breach October 2, and fires staff over mishandled data; Altman and Amodei urge slowing development. Hyperscaler AI capex tops 102% of operating cash flow; Temporal raises $550M at $12.55B, EliseAI $350M at $4B, Crusoe confirms $3B+. GitHub trending is all agent tooling and skill-security scanners (NVIDIA SkillSpector); Soup fine-tunes an 8B model on a 4GB GPU. The throughline: autonomous agents are creating real-world authorization failures faster than the industry can build controls.