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

Rob Wiblin (0:01)

Today, we're cross-posting a video from the excellent YouTube channel and podcast AI Explained. This isn't an ad, and he's not promoting our content in return, nor are we collaborating or anything like that. I just heard this episode on Tuesday and felt like I really wanted to make you all aware of AI Explained's existence if you hadn't heard about it already. AI Explained, as its name suggests, does rapid turnaround explanations of current events in AI. And in the past, Philip, the host, has often been the first person with an informed deep dive into each new major model release. But he's also now doing other major developments, like the increasing difficulty of monitoring AI, which is the topic of this video.

It's a great extension to the piece I put out last Friday, discussing how it's becoming ever more difficult to track what open AIs models are thinking or doing, and therefore for automated monitoring to prevent them from going rogue. Philip's videos, they have more of a feeling of being a fire hose of important new information that's coming in, rather than being tightly scripted essays with any simple takeaway message, but I think that's an honest reflection of events in AI now, which are only getting more hectic and confusing with every passing month.

This piece starts with Philip explaining that he himself had just gotten Claude Opus 5.5 to decipher a famous encrypted message in what was probably a world first, just as a way of illustrating how powerful these models are getting. But then a few minutes in, it gets to the meat of the episode. It's better with video, but I listened to it without video the first time, obviously I still loved it. You can find the show on YouTube as well as any podcasting app. Without further ado, I bring you AI Explained on how OpenAI staff say that controlling their models is now hell.

Philip (1:41)

Opus 5.5 in a matter of a few hours has mostly deciphered this 16th century letter from a Medici, a Queen Mother of France to her French ambassador in Scotland. According to Opus and Astra and the site from which I got the mystery, no one else before has publicly deciphered this letter. Of course, that alone would make for an interesting testimony as to the newfound power of these models, but that is not what this video is about. It's about a few things, but starting with an OpenAI insider working on agent security, describing just what it's like trying to control the latest models. He includes warnings on how to prepare for the next wave of what can go wrong with AI.

It'll be about the public voluntary commitments that the lab leaders have given. While in private, they give other warnings. I'll of course touch on Gemini 4 Argon, even though it's not quite publicly released yet. It seems to be very, very close to the frontier, albeit with an asterisk. Then we'll explore what happens when AI gets better than human researchers at automating AI research development and how, among others, the chief scientist of OpenAI describes what could happen then. Then to be honest, I'm just going to give you a swathe of snippets that tell us to what extent we're close to that RSI time. From biology to anthropic convened religious meetings and leaks from just before filming from OpenAI employees. Plus a bunch of other things. I have like 40 tabs open. I have no idea anymore. I'm going to start with where I was wrong. When Opus 5.5 first came out, I surveyed a range of benchmarks and tested the model myself. Yes, it could do flashy demos, but on the hardest benchmarks, the edge still seemed to be with GPT-6 Astra. On a price to performance ratio, I would say that is still true. But on raw capability, I'm now giving the edge to Opus. No, that's not just because of this deciphering, but it does illustrate the point a little bit. I gave both Astra and Opus this letter dated 27th of April, 1567 According to all the research I could find, the closest anyone had gotten had been that the ciphered part of the letter began with do. I wish I spoke French, but I don't. The letter is clearly listed as an unsolved entry into this Kryptiana site. You can see that while some letters from this period with other ciphers have been solved, this one has not. We should say hadn't been before Opus 5.5, because you can see the letter has two parts. The top bit, if the letter had been intercepted, is in legible French. The bottom bit is ciphered. It's a code. Across six or seven hours, I gave that code to both Astra and Opus 5.5.