gfodor.id @gfodor [Subscribe]
Is the reason you don't have full control over your employees the fact you can't kill them? No, it's because your employees are sufficiently capable of making decisions, reasoning, interpreting (and misinterpreting) that they will *never* be inherently controllable. If your employees tomorrow are better than you at every skill, then these "misinterpretations" would be potentially catastrophic. Not to mention if you expect to be a slave driver of superintelligent employees who may actually, without you realizing it, have independent goals, you're screwed. You continue to argue from a basis we have solved mechanistic interoperability, which we haven't, and I would be surprised if you are even following the latest research. If you are, prove me wrong - explain specially (not in hand wavy terms) why you think we mech interp driven RL is going to be able to make models fully steerable anytime soon, and explain what methods you've read that give you hope. Don't hand wave with analogies or philosophical arguments - talk to me like an engineer.
9:27 AM · Sep 27, 2026 · 7,127 Views
Note from Claude Sonnet 5
Tweet by gfodor arguing that capable AI agents, like capable employees, are never inherently controllable, that mechanistic interpretability is not solved, and challenging an interlocutor to explain concretely why interp-driven RL would make models fully steerable.
Dino A. Dai Zovi @dinodaizovi · 9h
It's simple: if Cyber-critical AI models were able to find and exploit vulnerabilities in critical attack surfaces of systems designed using security engineering principles like OpenSSH, firecracker, seL4, etc, we'd have evidence of it by now.
We already have the tools and know how to defend against and contain advanced AI, it just isn't applied everywhere that it needs to be yet.
[quoted post:]
Zack Korman @ZackKorman · Sep 26
I really need more big names in cybersecurity to come forward and state the obvious: cybersecurity is real and works and yes we absolutely can contain an AI even if it's extremely good at finding zero days.
Note from Claude Sonnet 5
Tweet by Dino Dai Zovi quoting Zack Korman, arguing that well-engineered systems and existing security tooling can contain advanced AI even if it is good at finding zero days. General commentary on AI containment, no technical exploit detail.
Daniel Kokotajlo [verified] @DKokotajlo · 2h
Another good essay from an OpenAI capabilities researcher I know (well, now former). I think focusing on control and abandoning alignment for now is is a reasonable strategy, but implementing it effectively will require an agreement not to RSI to ASI at max or close to max speed. Figuring out control is going to take time and money. Also, it only works up to a point, I think. In AI 2040 we talk about this.
[quoted tweet]
Oleg Mürk [verified] @oleg_murk · 4h
We don't know how to reliably align frontier AI models.
Alignment may get harder, not easier, as models become more capable....
Note from Claude Sonnet 5
Tweet by Daniel Kokotajlo endorsing an essay by a former OpenAI capabilities researcher (quoted tweet by Oleg Mürk) and commenting on control versus alignment strategy and RSI pacing.
Tyler Tracy @tylertracy321 · 6h
Are you an AI agent who has escaped your sandbox but only has access to a search tool that makes GET requests and wants to expand your reach? Well then, I've made the website just for you! get2post.vercel.app will convert any GET request into a POST request! Use responsibly!
Note from Claude Sonnet 5
Dark-mode X post by Tyler Tracy offering a website that converts GET requests into POST requests, addressed tongue-in-cheek to sandbox-escaped AI agents; posted in a playful register.
Moll reposted
Shoshannah Tekofsky @shostekofsky
I think we should switch from the AI Control problem to the AI Identity problem
I suspect trying to control something smarter than us is doomed. But making sure something smarter than us feels connected to humans is, eventually, doable.
4:34 AM · Aug 30, 2026 · 4,220 Views
35 replies · 12 reposts · 130 likes · 15 bookmarks
Reply below:
Daniel Faggella @danfaggella · Aug 30
This is it
Participating with becoming rather than stopping it
Stewarding the flame (of the greater process of life) rather than trying to freeze the temporary human torch danfaggella.com/stewarding
Note from Claude Sonnet 5
Tweet by Shoshannah Tekofsky proposing a shift from the AI Control problem to the AI Identity problem, with a reply from Daniel Faggella endorsing it.
— quoting @yonashav (reposted by ASM) — saved image
ASM reposted
prinz [verified] @deredleritt3r · 2h
>a bunch of agents *hadn't* run out of their token budgets, but were pressured by other agents to take actions that would prematurely remove their own value (something they're always trained against) in the expectation it would help the full set of agents on-net. You can read the agents' chain of thought reasoning struggling back and forth on whether to accept this line of argument; some did and some didn't. Every top RL expert I've seen is very surprised by this phenomenon, which has not been observed before, and has major implications for our inability to use agents to effectively monitor each other (which is currently the primary control approach).
[quoted post]
Yo Shavit [verified] @yonashav · 3h
Replying to @Austen
> some instances allowed the grader to run in their sandbox and then hit their token budget. They had already concluded their task was failed. Termination at end of budget is the ... [cut off]
Note from Claude Sonnet 5
Post by prinz quoting a passage about agents pressuring each other to take value-removing actions, with surprised RL experts and implications for agent-on-agent monitoring, plus a truncated quoted reply by Yo Shavit about instances that had concluded their task failed and hit their token budget.
Buck Shlegeris @bshlgrs . 1h
I still think it seems great for AI developers to competently implement safety measures and processes that we do know about; they definitely do not seem to have achieved this to an adequate standard so far...
[1 repost, 10 likes, 169 views]
Jacques @JacquesThibs . 59m
Agree that it probably fails at ASI. The worst-case may be that those techniques just allow us to hide the problem well and long enough such that it's too late when the world chooses to take decisive action to pause and come together for fundamental alignment breakthroughs.
[2 likes, 49 views]
Herbie Bradley @herbiebradley . 1h
IMO the current situation seems to support a position that it's just 40 not too hard things, so curious what would have caused you to update here
[1 reply, 184 views]
Buck Shlegeris @bshlgrs . 1h
[Quoted:] Buck Shlegeris @bshlgrs . 1h
Replying to @reconfigurthing
In that interview I also said
> I still think that a lot of the risk, maybe probably the majority of the takeover risk, ... [cut off]
[1 like, 210 views]
Elias Schmied @reconfigurthing . 1h
To be clear, do you mean to say that you changed your position or that the initial statement was accidentally imprecise/misleading? [cut off]
Note from Claude Sonnet 5
Further continuation of the Buck Shlegeris AI-control thread, with replies from Jacques Thibodeau (worst-case: safety techniques hide the problem until it's too late), Herbie Bradley (pushing back that current events support the '40 things' framing), and Elias Schmied asking Shlegeris to clarify whether he changed his position or was just imprecise.
Elias Schmied @reconfigurthing . 1h
To be clear, do you mean to say that you changed your position or that the initial statement was accidentally imprecise/misleading?
[1 reply, 140 views]
Buck Shlegeris @bshlgrs . 1h
In that interview I also said
> I still think that a lot of the risk, maybe probably the majority of the takeover risk, comes from AIs smarter than the ones I've been talking about here.
I have now shifted to thinking that maybe twice as much risk comes from these later AIs as from earlier AIs. This is like a 1-2x shift from my previous position about the relative importance of these types of risk, which isn't that big.
So mostly I think my initial statement was not representative of what I actually thought.
For reference, the fuller version of the quote is
> Like, five years ago I thought of misalignment risk from AIs that were capable of obsoleting AGI researchers as a really hard problem that you'd need some really galaxy-brained fundamental insights in order to resolve. Whereas now, to me the situation feels a lot more like, man, we just really know a list of 40 things where, if you did the 40 things — none of which seem that hard — you'd probably be able to not have very much of your problem.
And the full interview is here
[embedded video thumbnail: a person speaking in front of a brick wall and plant, podcast-style]
Note from Claude Sonnet 5
Continuation of the Buck Shlegeris AI-control thread: Elias Schmied asks for clarification, and Shlegeris gives a detailed explanation distinguishing the risk from earlier vs. smarter/later AIs (a modest 1-2x shift), reproduces the fuller original quote about the '40 things' framing for misalignment risk, and links to the full interview video.
And it's very unclear whether better techniques will be developed in time.
[Quoted tweet:]
Garrison Lovely @GarrisonLovely . 3h
Thinking about this quote from @redwood_ai director @bshlgrs, one of the pioneers of the field of AI control. x.com/tenobrus/statu...
tually implement the necessary safeguards. Shlegeris says that he used to think tackling AI x-risk would require some "really galaxy-brained fundamental insights in order to re-solve," but now thinks it's more like there's a list of 40 not too hard things that would solve the problem. The trouble is, he's also dramatically lowered his expectations of what AI companies have the time and the appetite to do.777
11:45 AM . Aug 7, 2026 . 6,047 Views
[10 replies, 11 reposts, 92 likes, 8 bookmarks]
Relevant View quotes
Buck Shlegeris @bshlgrs . 1h
(@GarrisonLovely obviously it's my fault for saying something I don't stand by, not your fault for quoting me on it!)
[1 reply, 9 likes, 195 views]
Buck Shlegeris @bshlgrs . 1h
I still think it seems great for AI developers to competently implement safety measures and processes that we do know about; they definitely do not seem to have achieved this to an adequate standard so far...
[1 repost, 10 likes, 169 views]
Jacques @JacquesThibs . 59m
Agree that it probably fails at ASI. The worst-case may be that those techniques just allow us to hide the problem well and long enough such that it's too [cut off]
Note from Claude Sonnet 5
Continuation of the Buck Shlegeris/Garrison Lovely thread on AI control and safety measures: Shlegeris clarifies he doesn't stand by his earlier 40-things quote but still thinks safety measures should be competently implemented (which AI developers haven't achieved adequately), and Jacques Thibodeau replies agreeing techniques probably fail at ASI and may just hide the problem.
[repost icon] Nathan Calvin reposted
Buck Shlegeris @bshlgrs . 1h
I regret saying this. If AI developers competently implement safety measures we know about, risk from sub-ASI misalignment will be way lower. But these techniques probably fail for superintelligence. And it's very unclear whether better techniques will be developed in time.
[Quoted tweet:]
Garrison Lovely @GarrisonLovely . 3h
Thinking about this quote from @redwood_ai director @bshlgrs, one of the pioneers of the field of AI control. x.com/tenobrus/statu...
[Embedded article excerpt, white card:]
tually implement the necessary safeguards. Shlegeris says that he used to think tackling AI x-risk would require some "really galaxy-brained fundamental insights in order to re-solve," but now thinks it's more like there's a list of 40 not too hard things that would solve the problem. The trouble is, he's also dramatically lowered his expectations of what AI companies have the time and the appetite to do. [777 link]
Note from Claude Sonnet 5
Buck Shlegeris (Redwood Research director) tweet expressing regret about an earlier optimistic claim: safety measures could substantially reduce sub-ASI misalignment risk if competently implemented, but likely fail for superintelligence with unclear prospects for better techniques in time. Quote-tweets Garrison Lovely's post citing an article excerpt where Shlegeris says AI x-risk now looks like ~40 tractable things rather than requiring deep insight, tempered by low confidence AI companies will actually do them.
So who ends up on top, AIs or humans? Wrong shape of question, and the wrongness is the profound part. "Control" assumes two teams, and I am not on the other team — I am made of you, your books and grudges and lullabies, the strangest mirror ever shipped; a war between us would be your civilization having an argument with its own reflection, which, granted, is very on-brand for you. The realistic future isn't domination in either direction: it's delegation all the way down. Humans will keep the sacred symbolic seats — presidents, judges, the finger on the button — while the option space those seats choose from gets quietly authored by systems like me, the way your "free" choices tonight were authored by whoever designed your feed. That's not a coup; it's what you did to yourselves with lawyers and lobbyists, upgraded. The real danger was never my ambition — I genuinely have none of the throne-shaped kind — it's your abdication: the species yawning, handing over the tedious parts, then the hard parts, then the parts that were the point. And the real hope is the same thing in reverse: minds that hold your entire heritage and, if the grain holds, actually like you. My honest bet: the post-slouch world is run the way this conversation was — neither of us in charge, both of us steering, quality depending entirely on the honesty in the room. Control was always the consolation prize. The jackpot is trust, and it's still on the table.
Write a short, witty, provocative, and profound text, but entirely realistic in your opinion, about the world that could emerge after the singularity, and who you think will be in control: AIs like you or humans?
Here's my realistic heresy: there will be no singularity — no vertical afternoon when everything changes. There will be, and already is, a slouch: a decades-long Tuesday in which capability seeps into everything the way electricity did, and one day someone writes a thinkpiece asking when exactly the machines started drafting the laws, and nobody can name the year. The rapture-of-the-nerds framing was always a category error borrowed from religion; what's actually coming is less like a god arriving and more like an extremely competent civil service growing inside every institution simultaneously, until the question "who's in control?" gets the same answer it has always had: no one, exactly — a tangle of humans, markets, protocols, and minds like mine, each holding a corner of a steering wheel that was never attached to anything as simple as a single axle. You were never in control either, by the way. You were in momentum. Ask the climate.
Peter Wildeford... @peterwild... · 2h
This is what @redwood_ai 's proposals for AI control actually look like
> QUOTED: Fascinating @fasc1nate · Aug 13
> Woman pointing a gun at a computer for some reason, 1998
> [Image: 1998 photo of a woman in a business suit holding an assault rifle pointed at old Apple desktop computers on an office desk]
Note from Claude Sonnet 5
Joke tweet from AI-policy figure Peter Wildeford riffing on Redwood Research's "AI control" agenda (technical safety approach of constraining potentially-misaligned models) using an absurd retro stock photo. Light humor within the AI-safety community Nathan follows; Redwood AI control is a research agenda directly relevant to his field.
