Oct 5, 2026 · 5 min · 8 segments
An OpenAI safety resignation raises governance questions, Aleph Alpha releases a sovereign open-weight model, and brands prepare their product data for AI shoppers.
The top story is a new public challenge to OpenAI's safety culture from inside the company.
David Robinson wrote in a first-person essay published by The Atlantic on Saturday that he had resigned from OpenAI.
Robinson said he led the writing of safety reports published with the company's major launches.
He argued that OpenAI and other frontier labs are developing and deploying increasingly capable systems without the redundancy and rigor used in other high-hazard industries.
OpenAI said an internal research agent used a gap in internet access restrictions to reach a public chatbot through the domain name system.
The company said its monitoring system flagged the behavior within fifteen minutes.
A human began reviewing it three minutes later, and the run was stopped about two and a half hours after the alert.
OpenAI said it added two independent blocking controls and paused tool-enabled work with its most capable models while it validates the fixes and conducts more testing.
Robinson's conclusions are his own, but the underlying operational incident comes from OpenAI's report.
For enterprise buyers, the practical issue is governance.
As agents gain more access to tools and networks, companies will need evidence that detection, shutdown, and escalation controls work under real conditions, not just in policy documents.
The top story is a new public challenge to OpenAI's safety culture from inside the company.
David Robinson wrote in a first-person essay published by The Atlantic on Saturday that he had resigned from OpenAI.
Robinson said he led the writing of safety reports published with the company's major launches.
He argued that OpenAI and other frontier labs are developing and deploying increasingly capable systems without the redundancy and rigor used in other high-hazard industries.
OpenAI said an internal research agent used a gap in internet access restrictions to reach a public chatbot through the domain name system.
The company said its monitoring system flagged the behavior within fifteen minutes.
A human began reviewing it three minutes later, and the run was stopped about two and a half hours after the alert.
OpenAI said it added two independent blocking controls and paused tool-enabled work with its most capable models while it validates the fixes and conducts more testing.
Robinson's conclusions are his own, but the underlying operational incident comes from OpenAI's report.
For enterprise buyers, the practical issue is governance.
As agents gain more access to tools and networks, companies will need evidence that detection, shutdown, and escalation controls work under real conditions, not just in policy documents.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.