The fastest way to understand what a driver-assistance system actually is, as opposed to what an ad says it is, is to read how the company describes it to its own investors. A risk-factor section is written by lawyers who do not benefit from optimism. That makes it the most honest paragraph a company publishes about its technology.
Tesla's annual report is a good test case. Its most recent Form 10-K, filed January 29, 2026 and covering the year ended December 31, 2025, discusses "alleged material misrepresentations in public filings regarding the effectiveness of Autopilot, Full-Self Driving (Supervised), and Robotaxi," and elsewhere describes accidents "where Autopilot, Enhanced Autopilot or FSD (Supervised) features are engaged." You can read the primary record on sec.gov; it was surfaced through EdgarBeast.
The load-bearing word is the parenthetical: Supervised. In the taxonomy the industry borrows from SAE, a Level 2 system performs steering and speed control but requires a human to monitor the environment and take over at any moment. "Supervised" is exactly that disclaimer, embedded into the product name. The car does a great deal; the driver remains the fallback and, legally, the responsible party.
Notice also what the filing is anxious about. It is not anxious that the feature does not work; it is anxious about the gap between how the feature is marketed and how it performs, and about accidents that occur while it is engaged. That is the recurring shape of ADAS disclosure across the industry: the technical capability is real, but the liability lives in the handoff between machine and human attention.
For a reader trying to separate spec from spin, the lesson generalizes. When a 10-K names a feature with a careful qualifier and then spends paragraphs on litigation around its "effectiveness," the qualifier is not boilerplate. It is the company telling you, in the one document where overstatement is punished, exactly where the system's limits are.