Unsupervised Full Self-Driving reaches customer-owned Teslas in Q4 2026
I’m just guessing here, but probably in the fourth quarter.
Elon Musk, 2026-04-22 — Tesla Q1 2026 earnings call — answering when unsupervised FSD reaches customer cars
Earnings call video
Electrek writeup with the line
Sealed at 12%
Locked 2026-08-31. Confidence 71%. Uncertainty ±7 pts. Resolves 2026-12-31. Status: PENDING.
How the engine got there
This is the nearest-term statement on the sheet and the one where the historical record is most directly informative. A version of "full autonomy is roughly a year away" has been stated publicly in most years since 2015, and each has passed without customer cars becoming unsupervised — close to a decade of consecutive misses on a single repeatedly restated proposition. A forecaster who discarded that base rate here would not be calibrated, merely credulous. The counterweight is real and we weight it: by May 2026 Tesla was operating vehicles with no occupant and no safety monitor in three Texas cities, which is a physical demonstration rather than a staged one, and genuine evidence the technical gap has narrowed further than at any earlier point. But the statement concerns CUSTOMER cars, a different regime. A company-operated fleet in chosen cities lets the operator select routes, conditions and liability exposure. Removing supervision from privately owned vehicles distributes risk across every road and every jurisdiction, converting the remaining question from an operational one into a regulatory and insurance one, and regulators have signalled nothing on this timescale. We also weight the speaker’s own hedge, which is unusually explicit — "I’m just guessing here" — and score the sentence as the guess it is presented as.
What counts as correct
CORRECT if, on or before 2026-12-31, Tesla enables unsupervised operation — no requirement for the driver to monitor the roadway or be prepared to intervene — on customer-owned vehicles in at least one publicly available software release in at least one jurisdiction. Steelmanned: a geofenced release, a limited early-access cohort, or a single US state all resolve this CORRECT. Tesla-operated robotaxi fleets do NOT count; the question asked and answered was specifically about customer cars.
We are wrong if
at 2026-12-31 every customer-owned Tesla still requires an attentive driver under the terms of the released software, including where a Tesla-operated driverless fleet runs in parallel.
Causal chain
Company-operated driverless fleets became real in 2026
Vehicles with no occupant and no safety monitor now run in three Texas cities, under conditions the operator selects and inside a permissive state framework.
The capability question moves closer to settled than in any prior cycle.
Customer vehicles remove operator control of the risk envelope
Private owners drive arbitrary routes in arbitrary conditions across arbitrary jurisdictions, so the operational containment that makes a fleet deployment tractable disappears.
Approval shifts from an engineering decision to a regulatory and liability one.
Regulatory processes are serial and slow
Authorisation is granted per jurisdiction with independent evidentiary requirements, and cannot be parallelised the way a software rollout can.
A four-month window is short relative to the observed cadence of these approvals.
Historical precedent
2015-2026 — Tesla full-autonomy timelines
The most directly relevant precedent available: the same proposition, from the same speaker, restated across roughly a decade, with each stated date passing.
2018-2024 — Waymo geographic expansion
A driverless system that genuinely worked still expanded city by city over years, gated by regulation rather than capability.
2016-2022 — GM Cruise deployment forecasts
Repeated near-term commercial forecasts from a well-capitalised programme, each slipping on approval and safety-case timelines rather than driving performance.
Directly related seals
More on this desk
Browse the Technology & AI desk, the full projection registry, the calibration ledger, or the record by subject. Scoring is explained in how a prediction is made.