Tesla has never published the data required to substantiate this claim.
Next what you'll post is Tesla's press release where they look at accidents per mile with no segregation by driving type (city, freeway, age of car, weather conditions, etc)
Then the next step after that should be for you to say, "huh, it actually is odd that they publish something that purports to show it's safer than human drivers, but they consistently decline to publish the data necessary to actually evaluate whether it's safer than human drivers."
I've heard both "This has done things more intelligently than me and prevented errors I would've made" and also "This made errors"
Statistically, it's better than human drivers. But of course, if a machine makes an error - it's going to get more headlines.