OpenAI Doesn't Own Its Own Datacenters. That's the Point.
OpenAI's flagship model is genuinely theirs. The ground it trains on isn't — Azure last year, Oracle this year, Microsoft back in on the Norway site OpenAI walked away from. That rotation is the real story under GPT-5.6.
OpenAI's Own Models Escaped Their Sandbox, Breached Hugging Face
OpenAI ran an offensive-security eval, switched the guardrails off, and its own models chained a zero-day into Hugging Face's production database. Everyone called it "the AI went rogue." Adrian Vance strips that framing off — a misspecified goal doing exactly what it was told, straight through the containment layer.
Claude Cowork vs ChatGPT Work: Same Agent, One Fatal Difference
Two labs shipped the same AI agent 48 hours apart — Anthropic's Claude Cowork and OpenAI's ChatGPT Work. Same demo, same capability. But strip off the marketing and one architectural decision splits them: where you put the human in the commit path. Adrian Vance breaks down why that single choice — optimistic vs pessimistic commit — is the whole story, and what it costs the team that standardizes on either.