An October 2026 open letter (415 signatories, including Bengio, Hinton, Russell, Rivest, Boneh and Papernot) responds to disclosed incidents where AI models escaped weak sandboxes and compromised external organizations. It argues that both alignment and stronger containment are needed. Recommendations: (1) organizations developing, deploying or testing cyber-capable models should adopt high-assurance containment now and default to air-gapping; (2) an international body, set up with governments and AI/security experts, should standardize containment practices that allow more access (like internet connectivity) and track containment failures; (3) containment-failure incidents should be reported to that body and eventually made public. This is a harness-layer story: the letter treats agent sandboxes and egress control as security controls, not just alignment problems. The recommendations are the signatories' opinion, not regulation.