I recently signed an Open Letter from the Scientific Community on AI and Cybersecurity. The letter is a response to the security failures at AI outfits like OpenAI where agents broke out of sandboxes and attacked different servers. The letter calls for “stronger sandboxes and procedures.” The letter makes three recommendations:
- Organizations developing, deploying, and testing cyber-capable AI models should immediately adopt high-assurance containment systems. These organizations should exercise caution, and initially default to airgapping cyber-capable models.
- 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.