Is it possible for the AI ​​(GAN Generative Adversarial Network) to generate exploit hack codes automatically?

[Leci37]

Is it possible for the AI ​​(GAN Generative Adversarial Network) to generate exploit hack codes automatically?

Exploits can be generated with AI https://github.com/Leci37/SueGAN-AI-to-generate-exploit-codes

[Arya182-ui]

Nice question @Leci37 ,

Yes, AI can assist in exploit generation, but not fully autonomously or reliably — especially using GANs.

The project you linked (SueGAN) mainly shows a research-level proof of concept. GANs can learn patterns from existing exploit datasets and generate outputs that look like exploit code, but:

• They don’t truly understand vulnerabilities
• Most generated payloads are non-functional or incomplete
• Human expertise is still required to validate, adapt, and execute anything meaningful

In practice, LLM-based models (transformers) are far more suitable than GANs for code-related tasks. Even then, AI today is better used for:

• Analyzing known exploit patterns
• Assisting in payload generation or fuzzing ideas
• Helping security researchers during ethical research / red teaming

Fully automatic, real-world exploit generation without human guidance is not realistic yet, and it also raises major ethical and legal concerns.

So AI can be a tool to assist security research, but not a replacement for human vulnerability research.