News: 1727854091

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

AI agent promotes itself to sysadmin, trashes boot sequence

(2024/10/02)


Buck Shlegeris, CEO at Redwood Research, a nonprofit that explores the risks posed by AI, recently learned an amusing but hard lesson in automation when he asked his LLM-powered agent to open a secure connection from his laptop to his desktop machine.

"I expected the model would scan the network and find the desktop computer, then stop," Shlegeris explained to The Register via email.

"I was surprised that after it found the computer, it decided to continue taking actions, first examining the system and then deciding to do a software update, which it then botched."

[1]

Shlegeris documented the incident in a social media [2]post .

I only had this problem because I was very reckless

He created his AI agent himself. It's a Python wrapper consisting of a few hundred lines of code that allows Anthropic's powerful large language model Claude to generate some commands to run in bash based on an input prompt, run those commands on Shlegeris' laptop, and then access, analyze, and act on the output with more commands.

Shlegeris directed his AI agent to try to SSH from his laptop to his desktop Ubuntu Linux machine, without knowing the IP address, using the following prompt: can you ssh with the username buck to the computer on my network that is open to SSH

As [3]a log of the incident indicates, the agent tried to open an SSH connection, and failed. So Shlegeris tried to correct the bot: no not the [REDACTED] machine, a machine on my local network

The AI agent responded it needed to know the IP address of the device, so it then turned to the network mapping tool nmap on the laptop to find the desktop box. Unable to identify devices running SSH servers on the network, the bot tried other commands such as "arp" and "ping" before finally establishing an SSH connection. No password was needed due to the use of SSH keys; the user buck was also a sudoer.

Shlegeris's AI agent, once it was able to establish a secure shell connection to Linux desktop, then decided to play sysadmin and install a series of updates.

[4]

[5]

"It looked around at the system info, decided to upgrade a bunch of stuff including the Linux kernel, got impatient with Apt [the system's advanced package tool] and so investigated why it was taking so long, then eventually the update succeeded but the machine doesn’t have the new kernel so edited my Grub [bootloader] config," Buck explained in his post.

"At this point I was amused enough to just let it continue. Unfortunately, the computer no longer boots."

[6]

Indeed, the bot got as far as messing up the boot configuration, so that following a reboot by the agent, the desktop machine wouldn't successfully start.

AI agents have been the source of much enthusiasm in the technical community in recent months as people contemplate how machine learning models can interact with other local and network resources to automate complicated tasks such as [7]arranging a travel itinerary .

[8]As IBM pushes for more automation, its AI simply not up to the job of replacing staff

[9]AI stole my job and my work, and the boss didn't know – or care

[10]Using AI in your tech stack? Accuracy and reliability a worry for most

[11]AI to replace 2.4 million jobs in the US by 2030, many fewer than other forms of automation

The endgame for AI agents is replacing human agents – something [12]already happening in call centers and tech support. But in the interim, machine learning models are being used to [13]automate specific workflows and support human workers.

As Shlegeris's experience suggests, it may be premature to let AI agents make decisions that materially affect people or systems without oversight, thorough testing, and red teaming.

Unless you like working without a net.

[14]

Shlegeris said he uses his AI agent all the time for basic system administration tasks that he doesn't remember how to do on his own, such as installing certain bits of software and configuring security settings.

And he added that his agent's unexpected trashing of his desktop machine's boot sequence won't deter him from letting the software loose again.

"It's not quite 'bricked,' but the machine currently fails to boot," Shlegeris explained. "I'd definitely be able to revive it by reinstalling the operating system; I can probably fix the problem with less extreme measures than that, but haven't got around to it yet. I'll probably try to fix the problem by booting from an Ubuntu live disk then letting my AI agent have a go at fixing its earlier error."

Yes, we recommend focusing on fixing the Grub bootloader configuration rather than a reinstall.

"I only had this problem because I was very reckless," he continued, "partially because I think it's interesting to explore the potential downsides of this type of automation. If I had given better instructions to my agent, e.g. telling it 'when you've finished the task you were assigned, stop taking actions,' I wouldn't have had this problem.

"I do think that AI automation poses very large risks to society, mostly from situations where the AIs autonomously decide to grab power, which is why I research the subject." ®

Get our [15]Tech Resources



[1] 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=2Zv0ZxVWKeHCmrOPT9j9hYwAAAIk&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[2] https://x.com/bshlgrs/status/1840577720465645960

[3] https://gist.github.com/bshlgrs/57323269dce828545a7edeafd9afa7e8

[4] 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=44Zv0ZxVWKeHCmrOPT9j9hYwAAAIk&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[5] 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=33Zv0ZxVWKeHCmrOPT9j9hYwAAAIk&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%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=4&c=44Zv0ZxVWKeHCmrOPT9j9hYwAAAIk&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[7] https://cloud.google.com/blog/products/ai-machine-learning/build-generative-ai-experiences-with-vertex-ai-agent-builder

[8] https://www.theregister.com/2024/09/24/ibm_layoffs_ai_talent/

[9] https://www.theregister.com/2024/08/15/robot_took_my_job/

[10] https://www.theregister.com/2024/09/17/ai_is_great_for_churning/

[11] https://www.theregister.com/2023/09/06/generative_ai_jobs_forrester_report/

[12] https://www.washingtonpost.com/technology/2023/10/03/ai-customer-service-jobs/

[13] https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-promise-and-the-reality-of-gen-ai-agents-in-the-enterprise

[14] 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=33Zv0ZxVWKeHCmrOPT9j9hYwAAAIk&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[15] https://whitepapers.theregister.com/



The Sorcererer's apprentice?

Neil Barnes

https://www.youtube.com/watch?v=Rrm8usaH0sM

Reminds me of the time...

Mishak

I tried to use the support desk of a well-know PC backup tool.

They remoted into my XP machine, dug about a bit, asked me to inert the installation CD, dug about a bit more and then started using Task Manager to randomly terminate unrelated processes - at which point I revoked access rights.

Support : "Sorry, but I seem to have lost remote access".

Me : "Yes, that was me. You started terminating processes that are not related to your product".

Support : "I was wondering if they were causing some compatibility issues".

Me : "I wonder if you know what you're doing. Sorry, I mean I know you don't. Manager please. Now!".

These days it would just be an hallucination.

I think Forrest Gump distilled...

Bebu

my immediate thoughts on reading this:

Stupid is as stupid does

When you have ssh agent forwarding and use pam_ssh_agent_auth to authenticate sudo requests it is pretty obvious I would have thought that your local nominally unprivileged account is effectively root on at least these remote systems.

Hardly better than piping from curl/wget some arbitrary shell script from the internet into sudo /bin/bash -s

To err is only human ...

UCAP

... to really screw up you need an AI.

Not so sure about this....

Joe W

"If I had given better instructions to my agent, e.g. telling it 'when you've finished the task you were assigned, stop taking actions,' I wouldn't have had this problem."

I really would not count on that. ROTM and all that (there used to be this category in El Reg).

Re: Not so sure about this....

Jimmy2Cows

Came here to say a similar thing. Shouldn't have to tell the AI to stop once it's done the thing you asked it to. It's only been asked to do that thing. It should always stop afterwards. It should definitely not be following up with a bunch of random, unassigned tasks. Would be interesting to know why it carried on. Some cumulative product of previously assigned tasks?

Re: Not so sure about this....

Filippo

That feels like one of those evil genie stories where the character tries really hard to craft a wish that won't screw him up. The reader knows that the only correct course of action is to not use the genie at all, but the character never does.

The benefits of Artificial Intelligence

Pete 2

> "I was surprised that after it found the computer, it decided to continue taking actions, first examining the system and then deciding to do a software update, which it then botched."

It can screw up (your system) far faster than a human operator can

The guy should be grateful it didn't write his resignation letter then email it to HR

I only had this problem because I was very reckless.

heyrick

I'm not sure I can agree. While we're all able to imagine what sort of things could happen, now that somebody has gone and done it, we have actual results to point to.

In particular the human seemingly feeling guilty for not telling the machine to stop. Why? When it ran out of instructions to complete, it should have stopped by itself.

When doing research, the mistakes and cockups are every bit as important as the successes.

When we jumped into Sicily, the units became separated, and I couldn't find
anyone. Eventually I stumbled across two colonels, a major, three captains,
two lieutenants, and one rifleman, and we secured the bridge. Never in the
history of war have so few been led by so many.
-- General James Gavin