TL;dr Raw material is still there, backup routine was not working. No disclosure on what the AI was supposed to do.
post
I kinda half hope that they have to remake the entire show with a 10$ budget.
Can't blame AI for the stupidity of not having a backup.
I work with AI a lot.
It works on copies. Even if there's no backup backup, it's not allowed to work in the main git tree - worktrees only, non-destructive commits only if there's no user supervision, the list goes on.
It's the bare minimum you can do.
I just don't let it commit, ever.
While I'd love for this to be true, the TV channel has denied that an AI agent would be to blame for the lost footage.
Even if the agent did delete the show, it wouldn't be the one to blame as the agent is unable to think on its own.
Let's entertain the possibility of the agent deleting the show; to do so, it should have been granted writing permissions, at least, on the folder where the show was located and, once that was granted (by a human), then a human should have ordered it to do something. The agent then executed the prompt in the most efficient way to achieve the result (which is not necessarily the best way) and this was the result. So yeah, a stupid human playing a stupid game won a stupid prize.
Is "AI deleted it" the new "my dog ate it" ?
I would be so fucking mad if I worked on this and some producer just gave AI direct access to the only copy.
The article plainly states that the production company says it wasn't AI, but instead that it was human error that led to the data loss.
The AI-thing was merely a rumour.
they say specifically "inte är en ”AI-agent som agerat självständigt”" (double quote). which translate to an "not an "AI agent which acted independently". so depending if that counts as human error or not... still up to to debate. It probably was AI, just not someone shift+delete the wrong folder.
I do however blame the backup crew which definitely should've had their backups in place.
Let's not let the truth get in the way of a sensationalist headline. It only matters that you click.
Humans run AI. Even if it was deleted by AI, it's still the responsibility of the person running it.
i would say it's the fault of whoever thought it was a good idea to hook it up to their data. the workers are just doing what they're paid to do and trying to put up with literal insanity or risk termination.
In 1997 a company I worked for "lost" a month of fully rendered 3D video. Big deal you're saying, but 30 days of a rendering farm in 1997 was worth a couple hundred thousand dollars.
Somebody formatted the drive by accident.
This shit happens all the time, it's appallingly common. I guarantee some idiot somewhere in the chain of command said no we are not budgeting $1,000 for a couple RAID arrays or backup drives
Happened to toy story 2 as well
https://www.newsweek.com/lightyear-producer-credited-saving-toy-story-2-after-deletion-1680072
No backups? Maybe I can understand not having full duplicates due to the cost, but surely it was on a system with at least snapshots in place or something right??
But they can restore from backups. They backed up their data, right?
Good. The more thins shit happens the more people will rip relying on it and see it as the unreliable bullshit that it is.
I wonder how many similar stuff need to happens before people realize they still need to keep their brains on when using AI and know what they're doing.
Oof.
As someone who edits my own videos for YouTube, can't imagine how much work that was. Just gone. Poof.
Oh well. Don't use bullshit AI next time.
Whatever the agent was actually doing, the real failure is that it had destructive write access to the only copy of finished work. That's a permissions problem before it's an AI problem — no automation, human or otherwise, should be able to hard-delete production assets without a human confirming the destructive step.
The fact that the season is just gone also tells me there was no immutable backup or soft-delete tier behind it, which is rough for a broadcast workflow where the footage is irreplaceable. Agents are getting broad tool access by default right now, and this is the cautionary tale for giving them delete rights alongside read rights.
Permissions that ai can access and change lol
Get out of here loser
I'd assume they're using cloud, so permissions are scoped per service. For example, if they used aws, they could have S3 create but no delete, and the role has no permissions to IAM (aws equivalent of permission policy)
If they just have some linux server then they can manage with users and have write permissions restricted to owner of the file.
They probably just let AI assume Admin and let it do whatever because nontechnical people don't understand least permission
The headline is an example of a new grammar structure, the agentic passive voice.
He’s the show’s host. So no, its not an interesting choice, its the obvious choice.
The person you're responding to seems to be a repeated race baiter on here.
I'm reading in translation where he is credited with "leading" the program, which could be host or could be showrunner.
As showrunner, he'd have more call to be there. As host, even less.
"Someone majorly fucked up and also here's a picture of someone" is kinda not great journalism, regardless.
He’s the public face of the programme as the host. As in, when the show is mentioned, its his face people recognise. Its standard unremarkable journalistic practice. Paranoia is not a good look.
Can confirm, he is the host.
I don't mind that he's black, but he is cringe af as a television host. Not professional or eloquent at all. Not funny, but definitely trying his hardest to be so. Lame jokes that he laughs at himself, sometimes the only one who laughs, which is... uuughhh.
I think he is there to appeal to a certain subdemographic of youth, and that is definitely not me. 😬 I can imagine certain people like him for his performance though.
"Not professional or eloquent at all. Not funny, but definitely trying his hardest to be so. Lame jokes that he laughs at himself"
Sounds like the timeless qualifications for being a reality show host tbh
Well, we hold people to a higher standard on certain channels. A few select channels we expect a higher standard of entertainment because of their decades long history of quality. But uh, yeah. I guess this guy's an acquired taste.
That's cool that there are still standards for any channels there, lol. What's the subdemographic you think he's appealing to?
Being the host makes it very not likely that he was the one responsible for managing the footage since those are two different jobs. That makes it unfortunate, if not borderline-libelous, to put his face on this article and give the impression that he was the one responsible for the fuck-up.
According to sources for Aftonbladet, Mastiff has allowed an AI agent into the server. Then the AI agent has deleted the pre-cut season of ”The challenge” from the server, with no possibility to recreate the project.
Rogue my butt. Someone must've tasked the thing to cleanup or update stuff and got monkey pawed.
I'll take "things that didn't happen" for $500. The AI didn't delete anything somewhere that it couldn't access on its own.
Someone gave the AI writing permission on a folder (or even drive) that shouldn't have it and then asked it to do something that obviously went as expected. People too dumb and lazy to do certain things are in charge of doing said things with AI leading to interesting results.
If you actually read the article, it says the AI deletion is just a rumour. The production company denies it and says that the real problem was human mistakes combined with a lack of proper backup procedures.
Aftonbladet is a tabloid so it tracks that they would keep the headline even when it is completely unbased. I have zero respect for them.
Sounds like someone is trying to keep their job or save face. An employee probably used Google's search AI summaries/suggestions to either set up their backup systems or perform some sort of maintenance command. The best excuse they could come up with was "the devil AI made me do it."
How many times do people have to keep getting burned before they stop sticking their hand in the fire?
"Maybe it'll become less painful and stop smelling like burnt ass if I trust the fire implicitly."
Well, it is always different people.
This is about not having proper backups, not Ai & the answer to your question is lots of times, "I'm sure my next HDD will never die".
That didn’t happen. But I bet someone’s very embarrassed to admit their mistake. “It was a rogue agent!!”
Why not both? Easy enough to footgun yourself with a bad prompt (because they're far from safe) then (somewhat justifiably) blame the LLM (that shouldn't have had that much access).
There’s only one way the agent got that level of access, and there’s nothing “rogue” about it.
The intent of this story as I read it is to scare the readership into the ploy that the big players need regulation - and the subtext there is to keep out the small players so that the incumbents can have a mono/duopoly.
I didn't read and won't read slop articles about slop outcomes, but I find very ironic what you say, after the latest marketing campaigns by the 2 or something biggest "influencers" in the ai space saying things for attention such as:
-
I'm not incompetent, but we must be careful or that book from 50 years ago will turn true and AI will kill us all!!
-
I'm not incompetent, but AI is sooooooooo such very much smarter than me that I put it in a crib and it jumped out with a back flip and in the process hacked 5 companies
Meh, the scary in its various forms is IMO marketing and a distraction to try to make it to IPO without having to admit LLMs are failing to scale as hyped (looking less likely all the time). Regulation would have to be worldwide to stop open weights models eating their lunch and the US just doesn't have that pull any more.
How can a word interpolation machine do anything on it’s own? Do they mean some script that feeds LLM things and then does things based on what it spews back? Who’s responsible for that script?
“Rogue” is just the inherent inaccuracy in ML models.
ML has metrics and those metrics all define how bad they are at doing the thing.

all 13 comments