Laboratory for Information and Decision Systems

Researchers at MIT developed an algorithm called HardFlow that forced generative AI outputs to satisfy strict, non-negotiable safety constraints in simulated robotics and imaging tests, meeting the re…