True randomness can make software programs safer, help prediction models operate better. But for decades, it simply didn’t seem possible. That has now changed.

Yoshua Bengio, an acclaimed A.I. researcher, is growing increasingly concerned by agentic A.I.’s unchecked behavior.