New method aims to keep kids safe from illegal AI-generated content

MIT built a way to audit AI models for illegal content risks without ever asking them to actually produce it.