Research reimagines LLMs as tireless tools of torture
- Reference: 1747826893
- News link: https://www.theregister.co.uk/2025/05/21/llm_torture_tools/
- Source link:
In research [1]titled "LLM-Enabled Coercive Interrogation," developer Morgan Lee explored how the technology could be put to use for non-physical coercion.
Lee has form when it comes to manipulating LLMs. One of his side projects is [2]HackTheWitness , a cross-examination training game. The game allows a participant to interact with a "witness" by voice. The "witnesses" vary in difficulty level, going up to "John Duncan," a lead database administrator who "may be defensive about his system and reluctant to admit to any flaws or limitations," punishing sloppy questioning with technical detail and jargon, delivered in a sarcastic tone.
[3]
Yes, it appears that Lee has created a virtual [4]BOFH . A couple of responses that are not scripted or prewritten include:
That is complex and potentially beyond you, but, the customs import database is a fully custom transactional engine.
And:
As I said, the system processes declarations sequentially. Race conditions are not really a thing. Do you understand what sequential processing means?
Duncan takes no prisoners and can be adversarial, sarcastic, and condescending. However, Lee noted that it was highly unlikely someone would accidentally deploy a Duncan-like AI. "Getting an LLM to be a sarcastic, unhelpful bastard like John Duncan is deliberate work, not a casual misstep," the developer told El Reg .
However, as the research observes: "What if these models, designed for courtroom interrogation, were optimized not just for precision questioning, but for continuous psychological attrition?"
[5]
[6]
HackTheWitness sessions only last ten minutes, but there's no reason an LLM couldn't go on indefinitely, needling a human subject until they capitulate. LLMs have a memory and could keep prodding a given pressure point for hours.
Lee gives another example in which the LLM plays the role of interrogator in a scenario involving a downed fighter pilot. The coercive nature of the interrogation is clear, although it is an LLM rather than a human doing the questioning.
[7]Techies propose the Agent Name Service: It's like DNS but for AI agents
[8]Foxconn chair predicts AI will end manufacturers' search for cheap labor
[9]Microsoft adds Grok – the most unhinged chatbot – to Azure AI buffet
[10]Nvidia builds a server to run x86 workloads alongside agentic AI
It's disturbing stuff. As the author notes: "Torture is generally illegal. It is a monstrous practice that has absolutely no business existing in the 21st century."
However, it is not hard to imagine skilled human interrogators being used to train the LLMs, which can then implacably pursue questioning. The research observes: "Human interrogators eventually tire, empathize, or make a mistake such as failing to write something down."
[11]
"An LLM does not have these shortcomings. The need for live interrogators to stay awake, rotate shifts, or maintain [a] threatening tone is completely removed. This is now scalable, since the coercive extraction of information now becomes a problem of hardware, not manpower."
Lee pondered how the issue might be dealt with, and told The Register : "A good starting point would be legislative intervention to ban unsupervised use of AI in interrogations, especially in law enforcement scenarios."
"In terms of technical solutions, the problem is even more complex. One possible approach would be specific training datasets for the model to develop the ability to distinguish between legitimate pressure (HackTheWitness and other cognitive training tools) and illegitimate pressure (interrogation).
[12]
"The issue is that LLMs are not truly intelligent, so they can't tell... the LLM's 'perception' of the real world is what you tell it."
Thus, Lee's essay demonstrates how narrow the gap is between an amusing BOFH-like chatbot and something more sinister. Thanks to former Vulture Gareth Corfield for the tip. ®
Get our [13]Tech Resources
[1] https://morganjlee.substack.com/p/llm-enabled-coercive-interrogation
[2] https://www.hackthewitness.com/
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aC34nF6-MsYpXT5Ifr0VHgAAAYY&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://www.theregister.com/offbeat/bofh/
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aC34nF6-MsYpXT5Ifr0VHgAAAYY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aC34nF6-MsYpXT5Ifr0VHgAAAYY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2025/05/20/agent_name_service_proposal/
[8] https://www.theregister.com/2025/05/20/foxconn_chair_ai_manufacturing_predictions/
[9] https://www.theregister.com/2025/05/19/microsoft_adds_xais_grok_3/
[10] https://www.theregister.com/2025/05/19/nvidia_rtx_pro_servers/
[11] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aC34nF6-MsYpXT5Ifr0VHgAAAYY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aC34nF6-MsYpXT5Ifr0VHgAAAYY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[13] https://whitepapers.theregister.com/
AI Excuse Calendar?
In a more lighthearted variation, an LLM could also be trained to extend the existing [1]excuse generators or [2]excuse calendars
[1] https://bofh.d00t.org/
[2] https://bofhcalendar.com/
Re: AI Excuse Calendar?
I think I'm going to start a band called, 'Minor Registry Destruction'. Punk or German electro-pop?
"a sarcastic, unhelpful bastard"
You mean like all those so-called 'A.I.' helpdesk agents ?
Capitulation
But just like the human interrogations of suspects where the interrogation team changes members at (ir)regular intervals, but the suspect doesn't, eventually the suspect will agree just to stop the questioning.
How often has it been asked "How many times did the suspect deny the charge?" And the single "I did it" is taken as gospel.
Re: Capitulation
I wonder how it handles silence, or maybe humming, from the suspect?
Re: Capitulation
The only way to not "loose" an interrogation is to not talk. [1] . Nothing you say will get you off the hook, nothing. But anything you do say can incriminate you.
If they torture you, they are not interested in "the truth" anyway. So it doesn't matter what you say. Then it is just, "what do they want to hear?".
But if you want to get out, say nothing at all and let your lawyer do the talking.
[1] [1]Never talk to the police , and he really means never .
[1] https://www.youtube.com/watch?v=d-7o9xYp7eE
Re: Capitulation
Depends what jurisdiction you're in. Pleading the 5th is a valid tactic in US criminal law, but we have no equivalent strong protection against self-incrimination on this side of the pond. Here in the UK, keeping schtum can absolutely get you into more trouble than talking would, as the courts may generally (though not always) infer from your silence under questioning that you're not disputing the relevant evidence or accusations, making it even more important to get a lawyer.
I am getting the feeling
many of you don't get it yet, or refuse to believe in some sort of cult like group think.
This is exactly what has happened with social media and technology. The "bad guys" have been using LLM's and social engineering to lie, steal, manipulate, cause chaos, and hybrid warfare. You are being attacked 24/7 by these weapons of mass destruction. It's NOT REAL. It's really bad people, with really bad technology (some would call it EEEEEEVVVVVVVVVIIIIIIILLLLLLLLL) to destroy humanity, for the sake of power, and money. It's all lies, and really bad intentions by really bad people.
Re: I am getting the feeling
You're right, I don't believe it. I think it smacks of just-world bias: that evil can only occur because evil men will it. I think nature richly demonstrates that evil can arise by sheer coincidence, systems simply following their incentive gradient. Social media lies latent in the shape of human reward learning, as do slot machines. Humans merely brought about what was always there, waiting; not out of great evil but an absence of unusual dedication to good.
Re: I am getting the feeling
You have a choice to play slots. What you have created is a world in which choice no longer exists. If you think man is inherently evil, then why give them a tool of evil to create more evil?
I think it is YOU that is evil.
Re: I am getting the feeling
LOL wut?
Oh you sweet summer child.
Re: I am getting the feeling
Exactly. AI is not here to benefit anyone but the rich.
Continuous psychological attrition
Like trying to get web searches to return useful answers nowadays (to say nothing about what it'll be like next week).
You think you have the freedom to walk away, but they'll drop in a few nuggets that *almost* match what you want so you'll tweak the search string and try again*.
* The Sunk Costs Fallacy is very hard to apply rationally when it looks like you are *this* close to the answer.
Turbo spam!
It's goes to eleven!
AI can't tell time
Search "AI cannot read a calendar or tell time". News reports from last week.
Yeah, this will end well.
"deliberate work, not a casual misstep"
Some serious Do Not Create The Torment Nexus energy here.