Urban Autonomy Lab
Chapter 2 · EV112

Tesla Confirms FSD (Supervised) Availability in China While Fleet-Wide Approval Remains Gated by MIIT — 21 May 2026

Date collected: 25 May 2026
Summary

Capability "available" but not pushable fleet-wide pending MIIT approval (target Q3 2026); a worked example of the ex-ante-approval regime and forced "Supervised" terminology

Headline Finding

On 21 May 2026 Tesla officially confirmed that Full Self-Driving (Supervised) is available in China, formalising years of delayed and limited deployment. The confirmation is narrower than it sounds: what exists today is a supervised, limited deployment, not a blanket rollout, and full regulatory approval to push FSD to every eligible vehicle in the country has not arrived — Tesla is now targeting Q3 2026. The episode is a clean worked example of a national regulator gating an over-the-air-deliverable driving capability through pre-approval, in contrast to the US, where Tesla deploys equivalent software broadly without comparable prior authorisation. It also sits inside China’s distinctive governance regime, which bans “autonomous/intelligent driving” marketing language and treats defect-fixing software patches as recalls.

Detail

  • Sources: CnEVPost, CNBC, CarNewsChina, InsideEVs, Sherwood News, Gasgoo — all 21–22 May 2026; SCMP. Window: 19–25 May 2026.
  • The announcement (21 May 2026): Tesla stated FSD (Supervised) is available in China. Previously Chinese customers could access only Autopilot and Enhanced Autopilot (precursors), with limited FSD versions available to select users pending approval.
  • Regulatory status: full regulatory approval to push FSD fleet-wide has not been granted; Tesla is targeting Q3 2026 as the window in which that could change. China’s Ministry of Industry and Information Technology (MIIT) did not respond to Wall Street Journal requests for comment.
  • Pricing in China: one-time purchase 64,000 yuan (approximately USD 9,400); no monthly subscription option offered to Chinese customers yet.
  • Competitive context: Tesla enters the Chinese assisted-driving market years behind domestic players (BYD, XPeng, and others), which have raced ahead; the launch heats up competition with domestic EV makers (SCMP). Tesla is also expanding Chinese battery sourcing (Seoul Economic Daily).
  • Regulatory regime context (from prior trawls): after a fatal Xiaomi assisted-driving crash in March 2025, MIIT (a) banned marketing terms such as “autonomous driving,” “intelligent driving,” and “advanced intelligent driving”; (b) moved to classify software patches that address a defect as recalls, potentially requiring production/sales suspension (EV55); (c) convened roughly 60 carmakers on 16 April 2025 to reinforce compliance; and (d) is advancing draft safety requirements for SAE Level 3/4 systems targeted for 1 July 2027 implementation (EV43, EV89).

Governance Implication

Four points for the paper.

  1. Regulatory approval as a market gate is a distinct governance archetype the paper should contrast explicitly (Ch.9). China requires prior approval before an OTA-deliverable driving capability can be pushed fleet-wide; the US permits broad deployment without an equivalent pre-clearance, with NHTSA acting after the fact through the Standing General Order and recall machinery; the UK (AV Act authorisation, EV58) and EU (type-approval/GSR plus AI Act, EV98) sit between. Tesla being able to “make available” but not “push to all vehicles” in China is the concrete illustration of the pre-approval model. The Ch.9 regulatory-horizon comparison should place these four regimes on an ex-ante-authorisation to ex-post-enforcement spectrum.

  2. Terminology governance is a live legitimacy instrument. MIIT’s ban on “autonomous/intelligent driving” marketing forces Tesla to position the product as “Supervised” (Level 2). This is a direct regulatory intervention against the safety-misperception failure mode — the gap between what a driver-assistance system is marketed as and what it can do — which the paper treats under Legitimacy (links to EV65 on ODD disclosure and EV83 on the Tesla HW3/FSD class-action wave). Ch.5 and Ch.7 should treat truth-in-capability-marketing as a named, transferable instrument, with China as the leading worked example.

  3. The OTA-as-recall regime intersects directly with the Waymo evidence this week. Under MIIT, the very mechanism by which FSD is delivered (over-the-air updates) is subject to recall classification when it addresses a defect — a sharp contrast with the US OTA-recall practice exposed as inadequate in EV109, where a recall could be filed before a remedy existed. Ch.7’s OTA-governance instrument should compare the Chinese “patch-as-recall, halt-until-resolved” model with the US “voluntary OTA recall, self-certified resumption” model as two ends of a governance design space.

  4. The development is primarily a Ch.2 global-landscape marker — a major Western ADS entering the world’s largest auto market under a stricter regulatory regime than its home market — but its governance interest is almost entirely in the approval gate rather than the deployment itself. For city-level governance the transferable lesson is that national pre-approval regimes can constrain capability rollout in ways city governments in permissive jurisdictions cannot, which strengthens the paper’s argument that where national frameworks are permissive, city-level instruments must carry more of the governance load.

Sources

Thread