I think there are ways Amazon could approach this. 1) They can have people review recordings for items that the system is less than sure of and charge the cust

I think there are ways Amazon could approach this. 1) They can have people review recordings for items that the system is less than sure of and charge the customer accordingly. 2) They could ban customers from the store if they determine that the customer seems to be attempting to subvert the system. Customers already steal things from stores. In this case, the fact that you have to tap-in makes it a lot less of an anonymous shopping experience which would likely deter people from trying to cheat the system. If people are trying to cheat the system, it seems relatively easy to ban them (compared to a conventional store). Amazon doesn't need to solve everything. Today, if you claim that you didn't receive a delivered package, what does Amazon do? I'm guessing that they just send another. However, I'd also guess that there's some trigger noting, "this person seems to have way more packages stolen than we'd expect". At that point, you can take action like requiring signatures or shipping to an Amazon locker. Similarly, Amazon will accept a certain amount of loss combined with the ability to note over time what is going on. The automated system may not be perfect, but presumably humans can review recordings. Ultimately, the only difference between current retail and this is the assumed intent. If someone walks out of a normal store without checking out an item, it's assumed that they intended to steal it. If someone walks out of an Amazon Go without paying, you assume they intended to pay and the system failed. However, if someone is putting their back to goods and picking them up from behind, that's pretty suspicious looking. It seems like Amazon's surveillance is meant to capture way more than typical retail and that should allow them to find bad actors easier, even if it requires a little human review.