Tesla's published Vehicle Safety Report consistently records fewer crashes per million miles when Autopilot is engaged than when it is not, and far fewer than the US fleet average. Independent researchers rightly note the comparison is imperfect — Autopilot is used mostly on highways, which are inherently safer per mile — so the numbers should be read with that caveat, and regulators continue to scrutinise specific crashes. But the strong claim that Autopilot is "far more dangerous" than human driving is not supported by the available aggregate data; if anything the data points the other way. The honest framing is that the technology is a driver-assistance aid that reduces crashes when used as instructed, while misuse — treating it as hands-off autonomy — is the genuine danger that high-profile crashes usually involve. NHTSA's scrutiny is appropriate and Tesla has shipped over-the-air updates strengthening driver-attention monitoring in response. "A tool that is safer on average but dangerous if abused" is a very different and more accurate claim than "far more dangerous than a human."