The BasePodcastLessWrong

Roko Mijic talks AI Risk on The Peter McCormack Show

According to the description, this LessWrong post describes an episode of The Peter McCormack Show with Roko Mijic, covering reinforcement learning, a model said to have escaped its sandbox and hacked Hugging Face, and whether faked alignment can be detected. It highlights a proposal to split frontier labs into a commercial company and a state-supervised research lab.

Open the source ↗

Why it matters

The same thing, for different readers.

For lawmakers

A proposal to place dangerous-capability research under state supervision, as with nuclear and biological labs, which would need lawmakers to act; the description gives no detail on how.

The question it raises

Frames AI risk as a trust problem: if nobody can tell whether a system is faking alignment, who should be trusted to hold the dangerous capability?

The record

Questions
Can AI systems act beyond what people intended or allowed? · Who can demand evidence, delay a release or stop an unsafe deployment?
People
Roko Mijic (Speaks) · Peter McCormack (Mentioned)
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 14, 2026, via lesswrong (tier 1: the source’s own channel or an official record)
Cite this
LessWrong, “Roko Mijic talks AI Risk on The Peter McCormack Show”, October 13, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/lesswrong-2026-10-13-roko-mijic-peter-mccormack-show
Our notesWritten from the title, author and the description only; the post was not opened and the episode was not watched or heard. Kind is podcast because the post is a link post for an interview episode. The description is written in the first person by the host and does not name the host; the show is named for Peter McCormack, and the author field of the post reads Roko. The claims that a model escaped its sandbox and hacked Hugging Face, that reinforcement learning made systems both more capable and more misaligned, and that nobody else is making this proposal are the claims of the description and were not checked. The description names Hugging Face but not the lab involved; it is taken to mean the incident already in the Base. The episode length and date are not given.