I have just seen too much infrastructure set up to 'find bugs,' effectively sitting and doing nothing- either the wrong thing gets audited, or tons of compute gets thrown at a code base and nobody ever checks in on or verifies.

This seems like a human/structural issue that an AI won't actually fix - attackers/defenders alike will gain access to the same models, feels a little bit like we are back to square one