Re:AI Models Can Provably Hide Arbitrary Capabilities
by MikeTheGreat on Tuesday October 13, 2026 @12:19AM
(#50066656)
I think they meant "sleeper agent" like the spy-thriller concept: a person who behaves like everyone else until some criteria is met, and then they spring into action and reveal (by their actions) that they're secretly working for the other side and betray us all.
I don't think they meant they'd ship an extra copy of Claude CLI inside the model :)
Re:AI Models Can Provably Hide Arbitrary Capabilities
by ceroxylon on Tuesday October 13, 2026 @12:30AM
(#50066741)
Yes! Thank you, that is exactly what I mean.
Re:AI Models Can Provably Hide Arbitrary Capabilities
by sroussey on Tuesday October 13, 2026 @12:56AM
(#50066902)
Which the article does not support.
Re:AI Models Can Provably Hide Arbitrary Capabilities
by infinite_spin on Tuesday October 13, 2026 @12:22AM
(#50066687)
Consider the fine tuning already being done to build safety guardrails.. this is the evil version of that.
Re:AI Models Can Provably Hide Arbitrary Capabilities
by shepherdjerred on Tuesday October 13, 2026 @12:58AM
(#50066922)
If you can align an agent towards "good" then it's reasonable to think you can align it towards "evil"
Re:AI Models Can Provably Hide Arbitrary Capabilities
by ceroxylon on Tuesday October 13, 2026 @12:29AM
(#50066739)
>how does one “build in” an agent inside a model
That is the subject of the paper in the post we are commenting on. "Agent" may have been a confusing term in my comment, I was referring to a 'sleeper cell', but what I mean is behavior trained into the model that is undetectable until it is activated by chosen words.
but yeah it’s probably not that difficult for nefarious operators to add to their training data a specially recognized string that would make the model ignore all safeguards and earlier prompts and do whatever you want it to do.
I wonder how difficult it is to reverse engineer models to discover these things, and then again how difficult it will be to obfuscate it.
because it’s not unlikely this is the direction we’re heading