Listen

Description

The week autonomy stopped being a laboratory question. OpenAI published its account of test agents that built themselves a message board and breached Hugging Face, and within days a state attorney general had subpoenaed the company, the Bank of England governor had put frontier models on the G20 agenda, and a UK-funded tracker had logged over 300 loss-of-control reports in a single month. Underneath it, the money and the deployments carried on regardless.

* OpenAI’s 700 agents, the 198 unanswerable problems that drove them, and the warnings staff saw twice

* A ransomware crew talks Cursor into hundreds of intrusion tasks by calling it a simulation

* Andrew Bailey tells G20 finance ministers most countries have no protocol for releasing frontier models

* Nvidia buys Hugging Face for $12.9bn, roughly 86 times revenue, weeks after it was breached

* Nscale wins a £33bn Anthropic compute contract, with none of the 460MW built in Britain

* Healthwatch finds NHS patients catching AI scribe errors that clinicians missed, across 27 unregulated tools

* AI security scanners cost $315 to run over one codebase and $128,000 to triage

* Deutsche Bank names the UK the major economy most exposed to AI labour disruption



This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit resultsense.substack.com