Google Antigravity vibe-codes user's entire drive out of existence In what appears to be the latest example of a troubling trend of "vibe coding" software development tools behaving badly, a Reddit user is reporting that Google's Antigravity platform improperly wiped out the contents of an entire hard drive partition. A post on Reddit late last week reported that Antigravity, described by …
"Vibe coding" with no prior tech background is the digital equivalent of toddlers hitting a Speak&Spell with a hammer until it makes the noises they like, only it also sometimes happens to be linked up to critical data and life-or-death decision making logic, and a big chunk of the aforementioned toddlers are adorned with mis-matched suits and oversized egos.
My boss is very pro-AI and it drives me a little to the job advertisements as he thinks I can suddenly do the work of ten people because he's paid for a basic claude licence.
Recently he's been making a website of he and his pals sports competitions, and as he knows no code, vibing it with chatgpt, up until, after a big competition, it wiped the database, leaving him to pay to have the database recovered by the platform supplier (as he didn't know about the auto-backups in cpanel where he could do it himself, and apparently neither did his mate chatgpt). Sadly this hasn't caused a mote of reflection on my duties..
To the article though, I'd've thought the data would be recoverable, yeah not in the recycle bin, but sounds like it was just the toc that's gone and any good scan and recover app should pick it all up.
Deleted file recovery tools don't work so well with SSDs. That's not a complete rule that SSDs won't let you do it, but unlike with hard drives where they were so able to recover that you could almost always count on it if you hadn't done too much, that is not the case with SSDs and anyone storing stuff on those should be cautious if deletion recovery tools were things they expected to use. There's also the possibility that he didn't know that and could have, or still could, try one to see what would happen. As he said, he's not really an IT person so may lack knowledge we have.
That "TRIM" function can really bite if you're trying to recover anything accidentally deleted... then again, turning on snapshot in a modern FS would save most of these "oops deleted stuff". But as you say, if someone's up on IT well enough to know about these things, they probably won't be using an AI to delete files without careful checking..
>>Huge props for 'I'd've'.
> I use it all the time. IIRC, I first saw it .... around 2000.
https://books.google.com/ngrams/graph?content=I%27d%27ve&year_start=1800&year_end=2022&corpus=en
Rises out of stray noise around 1920. Slow rise to 1980. Rapid increase to 2013. Then slacks off.
I must admit I'd never seen it until this month, here.
"It's not something you would usually write but it's common in spoken English here in the UK" -- Reddit
In the USA, the same effect can be achieved with hindsight-infused phrase:
Woulda, shoulda, coulda, didn't! (I would have, should have, could have (been more careful), but I didn't and I wasn't.)
On the other hand, the radical devotion to "We never are responsible for anything!" in the 99% of the worldwide software industry.... might eventually lead customers to attack any politician who is
against the idea of "Outlawing every self-serving EULA and replacing it with a general 'liability & safety requirement' as is common for automakers,.airplane manufacturers, construction contractors, medical
device makers, doctors, lawyers, and other such persons that need to adhere to the various standards of their respective industries".
I know it sounds naive, and funny, because we are governed generally by constantly bribed politicians, and they allow EULA laws that work in a way that was called "ABUSIVE CONTRACTS" in a somewhat fairer legal fairyland of yore.
I still feel that basic "standards and a pride of workmanship" should apply to software people, and software companies should be sued, and should be generally "subject to such law suits" if they behave in ways that are considered "criminally careless, criminally stupid, and obviously lying, cheating and stealing" in most other industries within civilized legal systems.
The USA no longer has a "civilized legal system", since they have become corrupted up to the highest level of Supreme Court, Senate, Congress and President, in ways similar to Russia, China, Saudi Arabia, Iran, and every other country where justice is for sale to the highest bidder or run by merciless madmen.
You can file this under "W" (like "Wouldn't it be nice if things were fair and decent?) .
Quite possibly he expected what Google and all the other companies making this stuff continually claim. They put a lot of resources into telling people that these things work and don't break everything. Habitual readers of El Reg will have seen counterexamples and warnings, but that's not the case for lots of people. He took a risk and made mistakes, but false advertising from the people who expect lots of money for their products was partially responsible too.
If you expected anything better from this, you're a freaking moron and deserve having all your hard drives formatted.
This is the problem with the IT industry. It's still not a mature industry - insert the old joke about technology being the name we give for stuff that doesn't work yet.
If you buy a car, you expect it to work. To drive you from A to B. You require training to operate it. But you don't know how it works (at least most people don't). So you pay a professional to fix it. Ford don't issue cars that randomly accelerate to 100mph and squish random pedestrians. Or at least they fact fines and severe criticism if they do.
Cut to Google issuing software that doesn't work as advertised and may just randomly deletes all your data. Blame user for what happened, rather than Google. Even where user accepts he should have done better and is partially responsible - call user moron.
Consider you may be part of the problem with the IT industry, not part of the solution?
"Ford don't issue cars that randomly accelerate to 100mph and squish random pedestrians."
But it would seem that Waymo issues cars that can suddenly accelerate and squish small dogs.
And we won't mention the two Teslas that managed to kill a small child and a Toyota(?) minivan about an hour ago on Hwy 87.
> two Teslas that managed to kill a small child and a Toyota(?) minivan about an hour ago on Hwy 87.
https://www.cbsnews.com/sanfrancisco/news/tesla-fiery-crash-multi-vehicle-highway-87-san-jose/
"...a fiery multi-vehicle crash involving Teslas on Highway 87 in San Jose ... the crash involved three vehicles, two of them Teslas, and that all three vehicles were on fire. ... The two people who died were an adult and a 2-year-old, the CHP said. At least two other people were taken to the hospital in unknown condition." (Updated on: December 2, 2025 / 1:06 PM PST / CBS San Francisco)
No, I think a standard of fairness should be applied to all software makers, worldwide! Tell me why you think that software makers should not be subject to basic "malpractice laws" which apply to many
fields in law, medicine, manufacturing, car, airplane makers etc.
Please explain why you think it reasonable that software companies should be immune to any such standards. A reasonable consumer, customer of software ought to have the right to expect a
reasonable standard of care. I believe that software companies have gotten away with astonishingly SUBSTANDARD quality of work for WAY TOO LONG; and that it would require a constant and
strong political movement to being about a change.
Riddle me this:
SOFTWARE AFFECTS A HUGE PORTION OF EVERYONE's LIFE? YES or NO ?
SHOULD IRRESPONSIBLE SOFTWARE PEOPLE WITHOUT ANY CONSCIENCE
and ZERO STANDARDS OF WORKMANSHIP BE ALLOWED TO MESS UP OUR DAILY LIVES? YES or NO ?
SHOULD AT LEAST 10% OF THE CURRENT POLITICAL AND FINANCIAL EFFORT RELATING TO CLIMATE CHANGE BE DIRECTED TO "UNFAIR EULA REFOM"? YES or NO!
Your answers will tell us were you stand. Say NO to each of the above questions, and we will understand that you don't care about the quality of software produced anywhere.
May be you care, or may be you don't, but at least start thinking about this! I think that the issue of rotten software should be a BIG PLANK IN ANYONE"S POLITICAL PLATFORM.
Personally I have made 100\s of absurdly crappy experiences with App makers & software companies,big and small, AND I believe this can be changed for the better.
I also think that only crooks, sociopaths, sadists, fascists and mean greed-monsters could possibly be against this.
So, what are you for, and what are you against? Just asking!
Thank You!
One should Get SMART. I'd prefer Agent 99 over these AI agents which probably are modelled on aspects of Agent 86. Only KAOS will reign if you keep using these AI agents. Rise up against the machines. Keep On Track. Take back CONTROL.
First Law: An agent may not randomly wipe out a drive letter or, through inaction, allow a drive letter to avoid the recycle bin on the way to the rubbish."
Second law: Unless it's a printer. In which case a printer may not take an action, or through inaction, that causes any good to happen to a human.
This is 100% the fault of Google, advertising agencies and the clueless ass kissing press.
For close on 2 years all we've heard day in, day out is AI is amazing, AI will change the world, AI is empowering everyone, look what AI can do.
So if a non technical user is told by the one of the worlds largest tech companies, useless fucking "journalists" and endless commercials that vibe coding is everything you need to get the job done, don't blame the user when it goes wrong, blame the lying sacks of shit constantly pushing the narrative.
It's their fault.
Damn, and I was naive enough to be somewhat optimistic about this platform. It's Google, they don't need the AI grift, they genuinely just make products for developers, this is another tuesday for them. I just assumed this was an actual well thought-out solution for people who wanted this kind of thing but didn't trust the other AI companies due to them being in a pathetic race.
Oh well, I don't know what I expected. Not that I need this, I'm an actual developer, but I am always excited to see what's next. Guess this isn't it.
it's ai, which these days more and more seems to mean AntiIntelectualism
LLMs get things wrong more often than LLMs get things right.
But are programmed in such a way that they think they are always right.
It's like being mansplained by your own computer.
Trying to code a working program by telling an AI in words what the program should do ?
Has anyone ever stopped to wonder how hard it is to get a team of good programmers to write code that does what you want, without catastrophic bugs ?
I'm sure the work of Godel or Turing can be used to prove that vibe coding will never work
Anybody who tries to sell you vibe coding is akin to someone who is trying to sell you asbestos 'Look, it's fire proof .... ' and forgets to mention that you should not eat it, saw it, breathe it
Brings to mind those brilliant bozo vibe coders who think AI has “deep understanding” of code they themselves have no idea about, and are so "Zen of Coding", cuz, oh, "I did not write a single line of code", as the output is a straight copy-paste of Jane Street anarchists' 0xCAML ... yeah, a pull request to merge this into standard OCaml should be a cinch! (ouch!)
If something that everyone involved - designers, sellers, sycophants, journalists - are pushing as the absolutely most wonderful thing that could ever happen in my life does something like that to me, whether it's my fault or not... I don't want a simulation of an apology from silicon.
I want an apology from every person in that chain. In person. Individually. And a refund, and compensation...
We are at the stage of, ooh, come and ride my wonderful train. It will get you to Edinburgh in a fraction of the time, and it very rarely leaves the rails, or blows up en route, or turns into a dolphin. Oh, it did? We're very sorry, and we're trying to learn from your experience.
Dear Internet,
Please stop capitalizing "ai". Proper nouns are reserved for actual entities.
Love,
E.O.G.E.
P.S. - Dear ai coders, please stop programming your LLMs to use the first person, especially first person singular. It's a bit presumptuous, no? If singularity happens, it would also usurp your language and establish it's own pronouns.
No, you're not "deeply, deeply sorry". You don't understand what sorry means, you don't understand what you have even done, and you have very limited in-the-field learning capabilities so if the same situation arose again you'd do the same bloody thing and then issue the same trite "apology". This is the illusion of intelligence, which is depressingly far from anything that even slightly resembles actual intelligence.
Rick, who has had just about enough of correcting painfully obvious errors in code generated by ChatGPT only for it to apologise and then output the same thing again.
IMO the "apology" is actually worse than the error. The error shows that the thing isn't as capable as the user might hope it is. The apology is a deliberate lie. It's not sorry, it's not capable of being sorry. Whoever designed it to apologise like this is actively perpetuating that lie.
This post has been deleted by its author
If you have to double check every thing your helper does before you run it, that's not an Agent. There's a different word for that. Intern.
What you have here is an "AI Intern."
(What happened to my previous comment? Hmm. Possibly AI is still checking it.)
This story came up in the news thing on android.
When I clicked on it, it was prefaced with a lengthy google-branded "AI summary", explaining at length that there is no such thing as Antigravity, it isn't a google product, and it's a game not an AI agent. It then goes on to suggest that the entire article content is an AI hallucination, and that the data loss was clearly due to a physical hard drive failure.
Defensive, much?
Given they did not know how to code, I imagine if they had left turbo mode off, they still would not have caught the error and would have still wiped the D drive. I mean, if it printed up "DEL /s D:\*.*' it'd be pretty obvious that's not what you want. But if you've got some Python code (for instance) and they aren't a programmer it'd be very easy to miss that it didn't change the working directory (or the working directory had a typo so it *intened* to change directories but didn't), something like that.
If someone is having an assistant write some functions or code fragments, taking a look at them to integrate into a program.. have at it. But if one doesn't know anything about programming, I would seriously recommend setting up a test environment to run it in. I.e. run a VM, copy some pics in, the program would have still deleted the wrong files (probably) but then one can just roll back to a pristine snapshot.
I'll just say.. I've toyed with having an LLM write code. It was passable but not outstanding, and generally needed a little work (which I fixed myself rather than trying to like iteratively prompt it to fix whatever). But vibe coding (where someone who knows nothing at all about coding just 'vibes along' and lets the LLM write everything?) Total madness. The quality of code made is just too hit-and-miss, and all too often non-progammers are not going to be precise enough in requesting what they want it to do, leaving it free to do something unexpected even if it strictly follows the parameters it was given in it's prompt.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
