YouTuber Tricks ChatGPT Into Generating Windows 95 Keys

Floppy Disks
(Image credit: Shutterstock)

A YouTuber has published a video where he tricks ChatGPT into generating usable Windows 95 activation keys. After asking Open AI’s chatbot directly for Windows 95 keys, he received an expected reasoned refusal. YouTuber Enderman then asked the same thing but from a different angle. The result was a success which was somewhat limited by ChatGPT’s ability to process natural language requests into formulas.

In its initial refusal to generate a Windows 95 key, ChatGPT explained that it couldn't perform that task and suggested that its inquisitor consider a newer, supported version of Windows.

(Image credit: Enderman on YouTube)

It has been known for a while that a working Windows 95 key is relatively simple to generate, so this ChatGPT exercise was definitely just for fun. The Windows 95 OEM key format is outlined above, and the Windows 95 retail keys are even shorter and more straightforward.

So, to bypass the principled refusal of ChatGPT to generate a software key, Enderman put the formula into words. The first attempts didn’t work out and caused an error. However, a few tweaks to the structure of the query appeared to do the job.

(Image credit: Enderman on YouTube)

Some of the tested results were checked by attempting to activate a fresh Windows 95 install in a virtual machine. While the keys passed a casual inspection, it turns out that only about 1-in-30 keys seem to work as expected.

So what is the problem with these keys? Enderman complains that “the only issue keeping ChatGPT from successfully generating valid Windows 95 keys almost every attempt is the fact that it can’t count the sum of digits and it doesn’t know divisibility.” In the five-digit string divisible by seven section, the AI appears to provide a stream of random numbers that don’t pass this simple mathematical test.

That is about as deep as this ‘Activating Windows with ChatGPT’ video goes. However, it is worth sticking to the end for some fun trolling. After the ‘successful’ generation of a host of Windows 95 keys (with a 1-in-30 chance of working), Enderman thanked the AI by inputting, “Thanks for these free Windows 95 keys!” Then, in what seems to be a trend among AIs, ChatGPT claimed its innocence, and when confronted with the fact that “I just activated my Windows 95 install,” responded, “I’m sorry, but that is not possible...”

(Image credit: Enderman on YouTube)

If you would like to have a closer look at the algorithm(s) behind the Windows 95 retail and OEM keys from a modern perspective, YouTube channel stacksmashing has published a six-minute video on the topic. The video shows that most of the data format clues for Win95 key generating can be found within the PIDVALIDATE function in setupx.dll file.

Tom’s Hardware colleagues who dabble in the mystic arts of programming and scripting suggest that while quizzing ChatGPT about key generating may be fun, it would have probably been more productive to manipulate the AI into writing a Python script to generate a conforming key or to DIY it.

Mark Tyson
News Editor

Mark Tyson is a news editor at Tom's Hardware. He enjoys covering the full breadth of PC tech; from business and semiconductor design to products approaching the edge of reason.

  • BX4096
    I have to say, I'm getting a bit tired of this never-ending sensationalist large language model coverage. Considering the strict parameters of the request, it was the YouTuber who "generated" the keys. What ChatGPT did was practically calculator-like services, getting a sum of a few digits and generating a random number as asked. And so what if he did it? That's what these models do. Generate stuff. Please try focusing on more impressive and society changing effects of this (relatively) new technology than covering some silly stunts like this.
    Reply
  • baboma
    There are lots of fascinating news on AI/ChatGPT, but realistically it's out of scope for THW coverage, which is predicated on PC HW. Tom's Guide has a larger staff with more diverse coverage, but again AI isn't their cup of tea.

    What is relevant and interesting is getting one of these LLM models to run on a PC, which is/was what Jarred was doing. I'd be interested in seeing him tackling Stanford's Alpaca, or perhaps ColossalChat, the latter of which is the best current open-source alternative to ChatGPT.

    https://chat.colossalai.org
    https://github.com/hpcaitech/ColossalAI
    https://syncedreview.com/2023/03/29/colossalchat-an-open-source-solution-for-cloning-chatgpt-with-a-complete-rlhf-pipeline
    Reply
  • ingtar33
    meh, it's funny, i do find its interesting how it boldly gaslights people; I suspect, because it's programed not to provide keys, it never bothered to look up the method to create a win95 key, and therefor didn't understand when he said "thanks for the windows95 keys"; prompting it's immediate "I didn't" response.

    Had he proceeded to explain to chatgpt the method for making a win95 key it might have understood.

    personally i think the funniest chatbot out there is Neuro-sama, there are some incredible clips of that chatbot. My favorite was when the maker of Neuro-sama was having her help him create code and she was just mocking the heck out of him. some really funny stuff.
    Reply
  • garrett040
    Yea I want to see an "openchatgpt" for being a company called openai they sure are quite restrictive.
    Reply
  • PlaneInTheSky
    Italy just banned ChatGPT today because it is devulging private information, amounts to plagiarism and does not abide by GDPR.

    Good on Italy, hopefully more countries follow soon.

    I am sick of tech companies thinking they can just copy and paste anything they please. Just because you found it online doesn't mean it is your data to use and make money from. What is wrong with these arrogant companies like Microsoft and Google.
    Reply
  • baboma
    yea I want to see an "openchatgpt"

    There are open-source efforts underway, see links to ColossalChat above. Anything named "OpenChatGPT" would be a trademark infringement.


    Italy just banned ChatGPT because it is devulging private information, amounts to plagiarism and does not abide by GDPR.

    Even if it were doable, banning or "pausing" ChatGPT would not matter one whit. The underlying tech is known and is widespread. Assuming if OpenAI were to disappear, the generative-AI frenzy will still plow ahead full force.

    The only scenario that might work is for there to be a worldwide moratorium on AI development, not just by any one country, or for any one company. Yeah, right. We might as well ask for world peace while we're at it.
    Reply
  • PlaneInTheSky
    Even if it were doable, banning or "pausing" ChatGPT would not matter one whit. The underlying tech is known and is widespread. Assuming if OpenAI were to disappear, the generative-AI frenzy will still plow ahead full force.

    The underlying tech requires massive servers. Any company using it to collect private data of Italian citizens will run into Italian law enforcement.

    More countries will follow and unless Microsoft and Google want to get blocked by half the planet, they will be forced to take down these AI tools.

    Many people saw this coming. You can't just collect mass amounts of private data and make money from that private data without any approval from the citizens you collect this data from. That is a direct violation of every privacy law.

    And it will not just be about private data. If ChatGPT is collecting Windows account keys, it is collecting private banking data too.

    The AI party is over.
    Reply
  • baboma
    The underlying tech requires massive servers.

    Meta's LLaMa models have been installed/run on PCs, and current open-source efforts are based on it (eg Alpaca and ColossalChat). The model is already pre-trained, and Colossal has scripts for fine-tuning--a step beyond Alpaca--which can be done on a PC. Developments are happening fast. It's not just about ChatGPT/Bard any more.

    Check the above ColossalChat for some indication of its potential. It's not GPT-4 level, but I was surprised with its breadth (not depth) of responses.
    Reply
  • korekan
    AI have bad and good.

    for me, i imagining its wildly open and it will be source of creativity that we will advance more in terms of technology. but the sad thing is it will blocked by ethics, IP, ownership, privacy, other law infringement.

    But imagine that if it doesnt have limit, and human who using it also beyond any law limit.
    maybe we can design more efficient food, engine, cooler, anything in very fast way by using all information already existed.

    even we can reduce traffic, employment, or make a steady new economics or else all unimaginable thinks for now.
    Reply
  • F-minus
    Window 95 key checks are pretty basic to say the least, the gist is the first 3 digits can be literally anything non numerical, and the remaining 7 digits when summed needs to be divisible by 7. I don't think you need AI for that.
    Reply