Our automated driving systems less safe than manually driving the car yourself?
Short answer: the evidence doesn't yet say Level 2 systems are less safe than driving yourself, and it doesn't clearly say they're safer either. It depends on the system's level, and the strongest numbers come from manufacturers.
For Level 2 (the category BMW's new feature falls in, like Tesla's FSD Supervised), a 2025 cross-methodological review in ScienceDirect found crash rates about 10% lower for cars with Level 2 engaged than for cars with only basic active-safety features. The same review warns that the benefit is smaller than for Level 1 features, which fits concerns about drivers becoming complacent. A separate analysis of one manufacturer's own crash reports (an engrXiv preprint, later indexed by TRID) found a claimed 43% advantage for Level 2 over Level 1 shrank to about 10% once it controlled for how much freeway driving each group did. Freeway miles are the easier, lower-risk kind, so raw comparisons flatter the system. The same analysis could not compare directly with general driving, because the manufacturer's crash-severity thresholds were unclear.
Driverless systems are a different question. An IIHS study of Waymo's robotaxis from 2021 to 2024 found a crash rate 68% lower than human drivers' in the cities studied, but Austin's sample was too small to be sure, and the robotaxis were involved in more rear-end crashes in an earlier naturalistic study.
These are mostly company-reported or small-sample studies, so treat them as suggestive. A useful test would be independent data on Level 2 systems on local roads, which is exactly what BMW's rollout will need.
