When it comes to AI-driven biorisk, the classic threat actor is a terrorist or deranged lunatic who would previously never have been able to produce a bioweapon, but will now, at some point, be able to do so. These are basically the people Dario was talking about in the biology section of The Adolescence of Technology (I know people had issues with this essay but I thought the biology section was quite good).

I actually think an underrated threat, relative to a malicious actor, could come from AI enabling more dangerous biological research, including gain of function experiments and work on mirror life.

1) Advanced AI will probably dramatically increase productivity in pathogen research, just as it will for other fields. I don't see any reason why people carrying out risky research won't also become much more productive.

2) As well as increasing the total number of experiments being run, very advanced AI could enable riskier research to be undertaken - for example, afaik there are many unsolved problems preventing us from creating mirror bacteria. Presumably AI will help us solve these.

3) The people who undertake this research are motivated by a desire to help prevent pandemics and think that this will help. Increased fear of biorisks, caused by AI progress (a 'Mythos moment' for biorisk), could actually increase the number of risky experiments that are undertaken.

4) Almost all of the safeguards that I've seen people propose to combat biorisks won't stop this. Since this work will be done by government or academia, they will have access to the latest models with biological safeguards removed, and obviously stuff like nucleic acid synthesis screening or restrictions on hardware won't do much.

5) The uplift required to enable this is much less than that required to allow some random lone wolf to create bioweapons on his own, so the risk will probably be here sooner.

NB: I'm not a biologist and have no special insight into the current state of the field.