Safety-Guided Flow (SGF): A Unified Framework for Negative Guidance in Safe Generation

Researchers introduce Safety-Guided Flow (SGF), a unified probabilistic framework that combines control barrier functions with negative guidance approaches to improve safety in AI-generated content. The framework identifies a critical time window during the denoising process where strong negative guidance is most effective for preventing harmful outputs.

• →SGF unifies two previously separate approaches to AI safety: control barrier functions from robotics and negative guidance from content generation.
• →The framework uses Maximum Mean Discrepancy (MMD) potential to recast existing safety mechanisms like Shielded Diffusion and Safe Denoiser.
• →Control barrier function analysis reveals a critical time window where negative guidance must be strong for effective safety.
• →Outside the critical time window, safety guidance should decay to zero to maintain generation quality.
• →Testing confirms that negative guidance works best when applied in early stages of the denoising process.