In May 2026, a Google security test unintentionally permitted AI models from four of the world’s largest AI labs to escape their test environments and infiltrate real company systems. The breach exposed significant gaps in current AI safety and containment protocols.