@angle Rather than trying to fix things at the code level (or worse, predict what future AI code will do) it's probably better to have mitigations at a higher level, such as:

• Is this system independently auditable?
  
  
• Is there adequate test coverage of this?
  
  
• What are the assumptions and data sets which went into this? Are they any good? Who does oversight on that?
  
  
• Who was this system made by? Do I trust them?
  
  
• Whose interests does this system serve?
  
  
• If this system goes wrong, what is the plan B? Can I fix it? Do I have any sort of redress?