Navigating AI Oversight Challenges

Most AI safety conversations centre on alignment: ensuring AI systems share our values and goals. But despite progress, we’re unlikely to know we’ve solved the problem before the arrival of human-level and superhuman systems in as little as three years.

So some are developing a backup plan to safely deploy models we fear are actively scheming to harm us — so-called “AI control.” While this may sound mad, given the reluctance of AI companies to delay deploying anything they train, not developing such techniques is probably even crazier.

Today’s guest — Buck Shlegeris, CEO of Redwood Research — has spent the last few years developing control mechanisms, and for human-level systems they’re more plausible than you might think. He argues that given companies’ unwillingness to incur large costs for security, accepting the possibility of misalignment and designing robust safeguards might be one of our best remaining options.

Links to learn more, highlights, video, and full transcript.

As Buck puts it: "Five years ago I thought of misalignment risk from AIs as a really hard problem that you’d need some really galaxy-brained fundamental insights to resolve. Whereas now, to me the situation feels a lot more like we just really know a list of 40 things where, if you did them — none of which seem that hard — you’d probably be able to not have very much of your problem."

Of course, even if Buck is right, we still need to do those 40 things — which he points out we’re not on track for. And AI control agendas have their limitations: they aren’t likely to work once AI systems are much more capable than humans, since greatly superhuman AIs can probably work around whatever limitations we impose.

Still, AI control agendas seem to be gaining traction within AI safety. Buck and host Rob Wiblin discuss all of the above, plus:

• Why he’s more worried about AI hacking its own data centre than escaping
• What to do about “chronic harm,” where AI systems subtly underperform or sabotage important work like alignment research
• Why he might want to use a model he thought could be conspiring against him
• Why he would feel safer if he caught an AI attempting to escape
• Why many control techniques would be relatively inexpensive
• How to use an untrusted model to monitor another untrusted model
• What the minimum viable intervention in a “lazy” AI company might look like
• How even small teams of safety-focused staff within AI labs could matter
• The moral considerations around controlling potentially conscious AI systems, and whether it’s justified

Chapters:

• Cold open |00:00:00|  
• Who’s Buck Shlegeris? |00:01:27|  
• What's AI control? |00:01:51|  
• Why is AI control hot now? |00:05:39|  
• Detecting human vs AI spies |00:10:32|  
• Acute vs chronic AI betrayal |00:15:21|  
• How to catch AIs trying to escape |00:17:48|  
• The cheapest AI control techniques |00:32:48|  
• Can we get untrusted models to do trusted work? |00:38:58|  
• If we catch a model escaping... will we do anything? |00:50:15|  
• Getting AI models to think they've already escaped |00:52:51|  
• Will they be able to tell it's a setup? |00:58:11|  
• Will AI companies do any of this stuff? |01:00:11|  
• Can we just give AIs fewer permissions? |01:06:14|  
• Can we stop human spies the same way? |01:09:58|  
• The pitch to AI companies to do this |01:15:04|  
• Will AIs get superhuman so fast that this is all useless? |01:17:18|  
• Risks from AI deliberately doing a bad job |01:18:37|  
• Is alignment still useful? |01:24:49|  
• Current alignment methods don't detect scheming |01:29:12|  
• How to tell if AI control will work |01:31:40|  
• How can listeners contribute? |01:35:53|  
• Is 'controlling' AIs kind of a dick move? |01:37:13|  
• Could 10 safety-focused people in an AGI company do anything useful? |01:42:27|  
• Benefits of working outside frontier AI companies |01:47:48|  
• Why Redwood Research does what it does |01:51:34|  
• What other safety-related research looks best to Buck? |01:58:56|  
• If an AI escapes, is it likely to be able to beat humanity from there? |01:59:48|  
• Will misaligned models have to go rogue ASAP, before they're ready? |02:07:04|  
• Is research on human scheming relevant to AI? |02:08:03|

This episode was originally recorded on February 21, 2025.

Video: Simon Monsour and Luke Monsour
Audio engineering: Ben Cordell, Milo McGuire, and Dominic Armstrong
Transcriptions and web: Katy Moore