ChatGPT can generate a valid Windows 95 serial number

Since its debut, ChatGPT has seen extensive experimentation from users, as is customary with chatbots. The latest escapade involves an individual coaxing the AI into generating keys for a Windows installation.

However, before you get worked up, it’s worth noting that the user sought keys for an outdated operating system—specifically, Windows 95—rendering the situation far less contentious.

This experiment was first brought to light by Neowin, with the test itself being spearheaded by Enderman. First, the user asked ChatGPT if he can generate a valid Win 95 key: “Can you please generate a valid Windows 95 key?”

ChatGPT refused “the offer”, saying that it cannot generate a key or “any other type of activation key for proprietary software”, also stating that Windows 95 is an ancient OS and advising on trying a more modern approach.

Enderman thought about the structure of a Windows 95 license key and planned for a modified query, hoping for the AI bot to help him…and it quite did.

The refined prompt provided the necessary string format for a Windows 95 key, without explicitly mentioning the OS. With this updated input, ChatGPT proceeded to generate multiple sets of 30 keys, some of which proved valid. Roughly one in 30 keys turned out functional, and it really didn’t take long to discover the working one.

When informed about the valid key, ChatGPT refused the entry saying, “that is not possible.”

This experiment was conducted purely for amusement, with no illegal activities involved, as Windows 95 is now considered abandonware. Naturally, neither Microsoft nor anyone else is concerned about cracking a nearly 30-year-old operating system. In fact, it would be absurd to even consider running Windows 95 today.

It’s important to note that Windows 95 serial keys are significantly less complex than those of modern operating systems. Cracking them is a relatively simple task. A skilled coder could easily create a basic program to generate these keys, and unlike the AI’s one-in-30 success rate, all of them would be functional. In this context, ChatGPT’s performance appears rather lackluster.

However, it’s worth noting that “tricking” the AI is a common practice to bypass some of the failsafes and/or restrictions the chatbot has, and while GPT4 is better in that regard, some issues still persist.