Technology · Earlier selection
A new algorithm checks AI’s final answers against hard constraints
MIT researchers have developed HardFlow, a method for steering generative models towards outputs that meet specified requirements.
Tests included robotic movement and image editing. The approach allows more freedom during generation, then enforces constraints on the result. Success in those experiments is not a general guarantee that AI systems are safe.
GO TO THE SOURCE
MIT News ↗ (opens a new tab)Primary institutional report or research paper, independently condensed. Findings and proposals are attributed to the source, with their limitations; no source prose, charts or photography reproduced.
AI-assisted original summary, checked against the linked source on 20 Sept 2026, 01:14 BST. Illustration is AI generated and does not document the event.
Report an error or rights concern