This research introduces a novel framework for selective content removal in Generative Adversarial Networks (GANs) using textual guidance, addressing critical ethical and security challenges.

• Enables targeted unlearning of specific content types without complete retraining
• Uses natural language prompts to guide the unlearning process
• Demonstrates effective content removal while preserving model quality
• Provides a pragmatic solution for AI service providers facing legal and ethical constraints

For security teams, this approach offers a practical mechanism to mitigate risks associated with generative AI by allowing selective content filtering without the computational expense of full model retraining.

Prompting Forgetting: Unlearning in GANs via Textual Guidance