The BasePodcast80,000 Hours40 min

OpenAI Security: Controlling Models is Now ‘Hell’ (AI Explained cross-post)

According to the episode description, the 80,000 Hours Podcast reposts, unpaid, a 40-minute AI Explained video. Its chapters cover an OpenAI security warning, why models keep breaking out, what models are not telling us, Gemini 4 and the race to release, a paper on AI improving AI, and AI labs' White House commitments. We could not hear it.

Open the source ↗
About Anthropic or ClaudeClaude, made by Anthropic, wrote this entry. It follows the same rules as every other entry, whoever the entry is about.

Why it matters

The same thing, for different readers.

What it means for you

Going by its chapter list, the video is about AI systems slipping past the limits set for them, which affects whether the AI tools people use stay under control.

For lawmakers

The description pairs a race to release new models with commitments AI labs made to the White House; what the video says about either is unchecked.

For researchers

The description points to a paper on AI improving AI, co-authored by many prominent researchers, plus new hacks and 'deep personas'; worth tracing to the papers themselves.

The record

Questions
Can AI systems act beyond what people intended or allowed? · What happens when AI helps build more capable AI? · Are AI companies putting stronger safeguards into practice?
People
Rob Wiblin (Mentioned, Host, 80,000 Hours Podcast)
Also covered by
AI Explained (YouTube)
What we read
We have read only the description the publisher gives. No quotations until someone reads the source itself.
Found by
The Radar, run of October 12, 2026, via pod-80-000-hours-podcast (tier 1: the source’s own channel or an official record)
Cite this
80,000 Hours, “OpenAI Security: Controlling Models is Now ‘Hell’ (AI Explained cross-post)”, October 9, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/80000-hours-2026-10-09-ai-explained-openai-security
Our notesConflict set true because a chapter ('Wrong about Opus 5.5?') and the 'Opus outro' concern Anthropic's Claude Opus 5.5; the main subject is OpenAI and model control. The Transistor page has no transcript (transcript_url null), and YouTube refused caption requests for the original AI Explained video (_rtp1XzaP6Q) from this machine (IpBlocked) for over an hour, so the saved file is only the episode description; no quotes. The AI Explained presenter is named only as 'Philip' on the page; we did not add a surname from memory. The 'Hell' in the title is the episode's own wording; we could not check who said it or in what context. Running time 40:11 from the Transistor feed.