AI physics lab Physical Superintelligence (PSI) announced progress on five long-standing quantum information problems using GPT-6 Astra and other models, including Claude, and released papers, proof code, and verification materials. According to ChainCatcher, physicists guided and checked the research, while AI models contributed to derivations and agents in Codex and Claude Code produced Lean formal proofs that computers checked.

One result shows that entanglement can persist between widely separated regions in a particular quantum field, although nonlocal correlations can no longer be detected using the CHSH Bell inequality beyond a certain distance. Another shows that repeatedly applying a certain class of quantum operations enough times will necessarily destroy entanglement. PSI said the five results do not fully resolve all the original problems: some findings apply only under specific conditions, and questions remain open.