They have said that the system won't be triggered on a single image. You would need to have multiple photos on your phone experience this hash collision which drops the odds of false positives considerably.

EDIT: It has now come out that you need to trigger the system 30 times before Apple acts on it. I can't imagine the odds for someone to have 30 hash collisions.

So they understand that their system is very susceptible to false positives, but they are saying their clients shouldn't worry because the black box hash gets compared with a black box inauditable threshold, both of which solely determined by Apple. I don't think the reaction was due to any confusion. People understood what Apple was trying to do and realized how much it sucked from a technical perspective.

Where did "very susceptible to false positives" come from? If the odds of a collision are one in a million that is troublesome if they only need one match. If they ignore anyone that has less than 3 matches, we don't really have to worry about false positives. People who have CSAM generally don't have only 1 or 2 images.

So why 30 matches before they do something? Why not 100? Why not 5?

I assume because they know the system is so false-positive-prone that even 29 false matches is a regular occurrence not worthy of wasting an employee's time to look at.