I wonder how long it will take the average Bay Area commuter to find an adversarial driving pattern for this.
Honestly I believe a system trained on so many millions of driving interactions would tend to act like an average good driver in SF, and so would ultimately blend in.