One of the easiest and most effective AI regulations would be to simply require all these labs to run all incoming requests through a suite of the most advanced models asking if the request is likely to be a pathogenic agent.
* there's not an obvious benefit to being the one nation that says "you know what? We want to be known as 'the nation whose biolab insecurity wrecked the world', because taking blame works out so well for everyone.", though I'm fairly sure a large portion of the world will be totally willing to wargame the impact of hosting a lab as a deadman switch so nobody else attacks them first…