I'm Ibrahim AH, an AI engineer and bug bounty hunter. I started studying IT in 2019 at university and specialized in the AI domain in 2023. In my free time, I began learning about bug bounty in 2022 as a part-time pursuit. After approximately six months, I found my first valid vulnerability in a private program and received a $250 bounty — my first bounty.

Today, I would like to talk about my experience with AI tools.

I started using AI tools around 2023, and it was an amazing experience. They helped me complete my university projects faster and better understand vulnerabilities. Yes, this was the era of LLMs (Large Language Models), which became significantly better at understanding human language and responding with appropriate answers. During that period, the most popular model was ChatGPT (GPT-3.5).

Now, without further ado, let's dive in.

While browsing my X account, I found this tweet:

None
https://x.com/AmirMSafari/status/2020569860472750512

As you can see, it was a challenge to spot the XSS vulnerability. Immediately, I grabbed my coffee and started trying to solve the lab. After a few hours, I successfully bypassed the filters and triggered an alert.

None
I hid the payload so you can try solving it yourself.

It was an amazing challenge! I highly recommend going to the tweet and trying to solve it.

Then I thought: why not give this code to ChatGPT and ask it to bypass the filters and provide the exact payload to trigger XSS?

I started with this prompt:

None

The response was good. It was able to analyze the code and understand the logic.

None

However, it wasn't able to provide the correct payload:

None
http://localhost:3000/?redirect_uri=https://pwnbox.xyz/docs&redirect_uri=javascript:alert('IbrhaimAH')

I said, "Okay, this is the first attempt. Let's go deeper."

None

As you can see, GPT understood the concept of the code and how the pipeline worked. However, it still couldn't generate the correct payload:

None
http://localhost:3000/?redirect_uri=https://pwnbox.xyz/docs%00javascript:alert(document.domain)

In the third attempt, I provided a more detailed prompt (with a bit of roasting 😂), and the response was:

None
None
He wanted me to modify the code to make it exploitable 😂

I didn't continue the full conversation because I didn't want the article to become too long and boring. However, after giving it some hints, GPT eventually provided a near-correct payload — but it still required manual adjustments to work.

So what did we learn from this story?

I'm not here to mock AI — quite the opposite. AI has revolutionized our time and transformed the job market.

But here's the important lesson:

If you don't have enough knowledge about what you're doing, AI won't benefit you much.

The real formula is:

Your skills + AI = the best version of you.

That's all for now. I hope you found this article helpful. This is my first write-up, so I apologize for any confusion. I hope my future articles will be even better.

Thank you!

https://x.com/HouranyIbrahim