AI Catchup — Saturday, November 7, 2026
Saturday, November 7, 2026Model
OpenAI releases GPT-6 Cyber for security research
A model built to find and validate exploitable bugs, with researcher-set guardrails.
ResearchAnthropic's AI research speed roughly doubled since 2025
Independent assessors found Anthropic's AI is now driving R&D progress at twice last year's rate.
ModelThinking Machines launches Inkling 2 Small, open weights soon
A 594B-parameter MoE model that can watch its own output and fix websites or 3D scenes as it builds them.
ToolsNvidia opens Cosmos 3 world-simulation models
Four open-weight models that generate photorealistic driving and robotics scenes from simulator controls.
PolicyCloudflare proposes trust framework for AI agents online
As AI agents browse and act for users, websites can't tell who they're really dealing with.