AI’s untestable reliability risk

Blankfein warns that modern systems can trigger massive losses quickly and argues regulators may need to slow deployment—not from sci‑fi fears, but because we can’t reliably test if these systems are right.