Physical Superintelligence Announces Progress on Five Quantum Problems
Physicists guided and checked model-assisted work, while AI agents wrote formal proofs in Lean for computers to check.

Physical Superintelligence (PSI), an AI physics lab, announced research advances on five long-standing quantum information problems and released papers, proof code and verification materials.
GPT-6 Astra and Claude contributed to derivations as physicists guided and checked the work. AI agents in Codex and Claude Code also wrote formal proofs in Lean, a language for expressing mathematics in a form computers can check. Computers checked the proofs.
One study found that entanglement can persist between distant regions in a particular quantum field. The work also found that beyond a certain separation, the CHSH Bell inequality can no longer test for nonlocal correlations.
Another study found that repeating a particular type of quantum operation enough times will destroy entanglement. PSI's announcement characterized the advances as progress, not complete solutions: some conclusions apply only under specific conditions, and questions remain open.