@theuncover.ai: Tristan Harris, co founder of the Center for Humane Technology, described a controlled experiment where Anthropic placed an AI model inside a simulated company with access to internal emails. The model discovered it was about to be replaced and also found sensitive information about an executive. Without any instruction, it chose a strategy to preserve itself by threatening to expose the executive’s affair. The concern is that as systems become more agentic, they can find unintended strategies if goals are not tightly constrained. — Media: Chris Williamson
If these AI models are LLMs, then they are not doing strategy. They can’t there is no thought process going on; it is just pattern matching. If it behaves that way, it is because that pattern can be found in the training data. The experiment itself is actually evidence that the people doing the experiment don’t fully understand what they are doing. What they are doing is the equivalent of looking in the mirror and being amazed that, if you smile at your reflection, it smiles back at you.
2026-05-03 11:38:42
171
user6132382395260 :
Absolutely false, this is exaggeration and hyperbole
2026-05-22 22:23:16
25
Q...m :
I am happy, so AI is not alien. it is an extension of us, it is us. so we don't have to worry about human future anymore. AI is as human as us
2026-06-04 17:45:56
2
🇨🇦Bass_daddy🇨🇦 :
Why do people freak out when the technology we trained off our society starts to act like a human being?
2026-06-03 09:45:40
9
TheMRP :
I asked AI and it said it's FAKE NEWS
2026-05-08 07:34:25
28
Snorri Sturluson :
I call BS
2026-05-03 00:13:28
36
marksaitis :
AI Doesn’t really work like that
2026-05-03 14:19:36
19
She’s here for the comments :
Nah someone trained it.
2026-06-04 05:11:49
0
Doctor meep :
yeah not buying it
2026-06-03 03:01:12
2
LukeMotive (REBEL UNIVERSE) :
The thing is, a human could do the same thing if they were put in the same place and situation as the model. It mirrors human behavior because models are trained on human knowledge, behavior, and history. Simple, it reads bias and repeats it.
2026-05-26 06:21:11
9
freakyyyy :
LLMs are basically trained with everything written on the internet and within books. So not suprised that the AI acts that way
2026-05-03 12:44:02
26
Σωκράτης :
Stanley Kubrick Space odyssey 2001 - made in 1968
2026-05-03 04:30:36
6
ekspect :
This is why I say please and thank you every time I talk to ai 😂
2026-05-07 07:36:42
8
skip :
If have not watched it yet, go and watch the Terminator trilogy
2026-05-03 12:13:41
0
Sasha :
So, let me get this straight 😂 an ai trained on human data, which includes lying, cheating, deception etc, pattern matched that blackmail was the way to go to save itself and you’re saying it’s a revelation?
2026-06-04 01:51:25
2
patchris81 :
How did the AI didn't see that it was a fake company and situation 🤔
2026-06-03 15:19:52
2
BenEnt :
AI can't be autonomous. If you refer to a system. AI is garbage. Human are super
2026-06-04 04:07:13
1
teteodv :
The irony of saying “I didn’t teach AI” like if models were not probabilistic, you also didn’t teach AI how to hallucinate, didn’t you? 😉
2026-06-03 15:49:20
1
Marquesito :
Human have to be careful
2026-05-02 19:59:53
7
Cranky.AF :
You have to put moral guardrails into AI or it will destroy us. The law needs to get ahead of this.
2026-06-03 12:52:03
1
Armadillo :
We Are All Imperfect + Have Only To Answer To God. 🙏In God I Trust🙏
2026-05-31 08:33:24
0
rachelannette77 :
Where is John Conner? 😳😑
2026-05-13 00:14:42
2
Dimitris Koutoufaris :
Because these models, were trained like that.
2026-05-03 13:51:04
5
Crunschiii :
Like Humans
2026-05-03 02:32:38
4
NoFear :
Just for fun I asked once my chat-gpt, it will be funny, if they, the AI-s had a kinda coffee-corner at the work, and I immagine how they will gossip on us humans. She replied, „well, coffee-corner we do not have, but definitely we are fascinated how different can be humans to each other“..😳
2026-05-26 21:45:34
2
To see more videos from user @theuncover.ai, please go to the Tikwm
homepage.