Skip to main content
ExplainerDriver MonitoringTrade-Off AnalysisAug 21, 2026, 5:50 PM· 3 min read

Tesla Recalls 2.7 Million Vehicles in China to Upgrade Driver Monitoring Systems

Tesla is rolling out an over-the-air software update to 2.74 million vehicles in China to shift its driver-assistance monitoring from steering-wheel torque to cabin-camera eye tracking.

By Anastasia Kuznetsova

Regulatory Authorities 40%Automotive Engineers 35%Privacy Advocates 25%
Regulatory Authorities
Argue that physical touch is insufficient for advanced driver assistance and mandate direct visual monitoring.
Automotive Engineers
Focus on the technical trade-offs between sensor reliability, computational load, and spoofing vulnerabilities.
Privacy Advocates
Raise concerns about the continuous recording of biometric data and cabin surveillance.

The short answer

  1. Tesla is recalling 2.74 million vehicles in China to upgrade its driver-monitoring system via an over-the-air software update.
  2. The update shifts the primary attention check from steering-wheel torque sensors to interior cabin cameras.
  3. Chinese regulators mandated the change, ruling that physical hands-on-wheel checks are insufficient for advanced driver-assistance systems.
  4. The recall highlights the industry-wide trade-off between the privacy of torque sensors and the precision of biometric eye-tracking.

Tesla is deploying an over-the-air software update to 2.74 million Model 3 and Model Y vehicles in China to fundamentally alter how it monitors driver attention. The recall shifts the primary safety check for the company's assisted-driving system from the steering wheel to the interior cabin camera.[1]

The sweeping action was mandated by China's State Administration for Market Regulation (SAMR). The regulator ruled that Tesla's existing mechanism for Level 2 "combination driving assistance" was insufficient to warn drivers when their gaze left the road, increasing the risk of collisions.[1]

For years, the automotive industry standard for driver monitoring relied on steering-wheel torque sensors. These systems measure physical resistance, requiring the driver to periodically apply a slight turning force to the wheel to prove they are engaged with the vehicle.[2][3]

The trade-offs between physical and visual driver monitoring systems.

However, torque sensors cannot verify cognitive or visual attention. A driver can keep a hand resting on the wheel while looking down at a phone or even sleeping. In China, some drivers actively bypassed the system by attaching aftermarket weights to the steering wheel, tricking the sensor into registering continuous physical contact while they rode hands-free.[6]

However, torque sensors cannot verify cognitive or visual attention.

To close this loophole, the SAMR recall forces Tesla to activate its interior cabin camera across the Chinese fleet. The camera, mounted above the rearview mirror, uses machine learning algorithms to track eye gaze and head position. If the system detects that the driver is not looking at the road, it issues escalating audio and visual alerts.[3][4][5]

While cameras measure visual attention directly, they introduce entirely new spoofing vulnerabilities. Shortly after Tesla began rolling out its supervised camera monitoring, Chinese drivers were filmed using $20 plastic doll heads mounted near the rearview mirror to fool the eye-tracking software into registering an attentive human face.

The updated software will issue escalating alerts if the cabin camera detects the driver's gaze leaving the road.

The transition also carries significant privacy implications. Tesla had previously resisted activating cabin cameras in the Chinese market, citing data security and local privacy laws. The SAMR mandate effectively forces a fleet-wide adoption of biometric surveillance, reflecting a global regulatory consensus that direct visual monitoring is necessary for advanced driver-assistance systems.[2][6]

The driver-monitoring update was one of two massive regulatory actions Tesla faced in China on the same day. A separate, overlapping recall covered 2.98 million vehicles—including imported Model S and Model X units—to address emergency mechanical door releases that regulators deemed too difficult to locate after a severe crash and power failure.[1]

Unlike traditional automotive recalls that require owners to schedule dealership visits for physical repairs, both of Tesla's fixes are being deployed remotely. The ability to resolve millions of regulatory safety violations overnight via software updates highlights how connected architecture is reshaping the mechanics of automotive compliance.[1][6]

The scale of Tesla's dual regulatory recalls in the Chinese market.

Competing readings

Torque-Based Monitoring (Hands-on-Wheel)

Relies on steering column sensors to detect physical resistance, ensuring the driver is physically engaged with the vehicle.

FOR: Highly reliable in all lighting conditions, requires minimal computing power, and preserves cabin privacy since no images are captured. It is entirely immune to driver accessories like polarized sunglasses or heavy winter hats. AGAINST: Easily defeated by aftermarket weights or simply resting a knee or hand on the wheel without looking at the road. It measures physical proximity, not cognitive attention. EVIDENCE: Historically the industry standard for Level 2 systems; however, Chinese regulators (SAMR) ruled it insufficient for Tesla's 'combination driving assistance' after widespread misuse. FITS WELL WHEN: Used in basic lane-keeping systems or as a redundant secondary check. DOES NOT FIT WHEN: Deployed as the sole attention check for advanced, hands-free capable Level 2+ systems.

Camera-Based Monitoring (Eyes-on-Road)

Uses interior cabin cameras and machine learning to track eye gaze and head position, verifying visual attention to the road.

FOR: Directly measures visual attention, which is a vastly superior proxy for cognitive readiness than physical touch. Can issue immediate, escalating alerts if a driver looks at a smartphone, turns around, or falls asleep. AGAINST: Computationally heavy, raises severe privacy concerns regarding continuous in-cabin recording, and can be obstructed by poor lighting or specific eyewear. It can also be spoofed, as demonstrated by drivers using plastic doll heads to trick the facial recognition. EVIDENCE: Tesla is deploying this via an over-the-air update to 2.74 million vehicles in China to comply with SAMR's stricter attention requirements. FITS WELL WHEN: Paired with advanced Level 2+ systems where drivers are highly tempted to disengage visually. DOES NOT FIT WHEN: Drivers wear heavy facial obscuration or operate in jurisdictions with strict biometric privacy bans.

2.74 million
Vehicles recalled for driver monitoring
2.98 million
Vehicles recalled for door releases
Level 2
Automation level of Tesla's system

Sources

Source coverage

6 outlets

3 viewpoints surfaced

Regulatory Authorities 40%Automotive Engineers 35%Privacy Advocates 25%
  1. [1]ForbesRegulatory Authorities

    China Orders Tesla's Largest-Ever Recall For Nearly 3 Million Cars

    Read on Forbes
  2. [2]WikipediaAutomotive Engineers

    Driver monitoring system

    Read on Wikipedia
  3. [3]AptivAutomotive Engineers

    What is a Driver-Monitoring System?

    Read on Aptiv
  4. [4]ValeoAutomotive Engineers

    Driver Monitoring System

    Read on Valeo
  5. [5]TeslaPrivacy Advocates

    Autopilot and Full Self-Driving Capability

    Read on Tesla
  6. [6]Factlen Editorial TeamPrivacy Advocates

    Synthesis by Factlen editorial team

    Read on Factlen Editorial Team

Comments

Stay informed

Every angle. Every day.

Get transportation stories with full source coverage and perspective breakdowns delivered to your inbox.