Comment by Nicolas Papernot

In consultation with governments and a wide range of both AI and cybersecurity experts, establish an international body that standardizes, for the public good, best practices for alignment and containment approaches that, unlike airgapping, allow for greater access outside the sandbox, such as internet connectivity. This body should also be mandated to record and track containment failures. To ensure that AI safety and containment research can learn from mistakes, security incidents involving containment failure must be reported to the body and eventually made public. This transparency is required for both learning and accountability purposes.