In April 2025, China’s Ministry of Industry and Information Technology (MIIT) issued a regulatory notice for intelligent connected vehicle products, pulling the automotive world’s handbrake on hype. This comes directly after a fatal car crash in late March caused by the incorrect use of assisted driving features.
For years, automakers and technology suppliers had been promising autonomous or smart driving, but real‑world performance often lagged behind the marketing.
New measurements shared by the MIIT are not just headline‑grabbing restrictions. They represent a fundamental shift from bold claims to verifiable safety.
Management teams working on advanced driver‑assistance systems (ADAS) in China must understand the new rules, adapt product roadmaps and supplier strategies, and prepare for a regulatory climate that values evidence over hype.
This article summarises the regulatory changes, explains their technical meaning, compares global frameworks, and analyses how companies like BYD are responding.
What does the Regulation Demand?
Advertising discipline and ban on inflated terminology
MIIT told automakers to stop using terms like “smart driving,” “intelligent driving,” or “autonomous driving” when marketing Level 2 systems.
Only explicit SAE‑level language, such as L2-assisted driving, is permitted. The Ministry of Public Security later signalled that overselling ADAS could violate China’s Advertising Law, which allows fines of three to five times the advertisement cost, or CNY 200 000–1 000 000 when the cost is hard to determine. Misleading promotion can also lead to suspension of business licences.
For management teams, this means that communication strategies must be cleared with legal counsel; marketing content, owner manuals, and dealer training must reflect the new nomenclature to avoid penalties.
Public testing and feature restrictions
Previously, some companies used pioneer user programmes to test new functions on public roads. MIIT forbade public beta testing without governmental approval. Demonstrating new features now requires either controlled conditions or special authorisation.
Functions perceived as convenience rather than safety—such as remote summon, valet parking, and remote parking—were banned outright because regulators concluded they cannot ensure continuous driver engagement.
This directive affects any features that allow the driver to leave the seat or that enable the vehicle to drive itself in confined areas. Thus, product teams must remove these functions from Chinese builds or redesign them for strict supervision.
Hands‑on requirements and driver monitoring
MIIT emphasised that Level 2 systems are assistive; the driver must remain engaged and be ready to take over at any time. The new rules require driver‑monitoring systems that detect if a driver removes their hands from the wheel or looks away.
If hands‑off detection exceeds 60 seconds, the system must issue warnings and then slow down or pull over. Later drafts of the national standard for Level 2 combined driver‑assistance systems reinforce these conditions. The draft (September 2025) emphasises human‑machine interaction, cybersecurity, and data recording.
It mandates hands‑off and gaze-monitoring, requiring the system to issue warnings and temporarily disable itself for repeated non‑compliance.
Yicai Global notes that the draft also calls for the vehicle to record key data for accident investigation and to disable itself for at least 30 minutes if warnings are ignored.
These requirements mean suppliers must upgrade steering‑wheel sensors, camera‑based driver monitoring, and analytics for logging behaviour.
OTA updates and recall procedures
One of the most significant changes is the way China now treats software updates. The MIIT & State Administration for Market Regulation (SAMR) notice, which entered into force on 28 February 2025, classifies OTA updates into categories.
General software improvements can be deployed after filing; updates affecting major technical parameters need regulatory approval; upgrades related to Level 3 or higher functions require admission approval; and OTA updates for recall fixes must comply with recall regulations, with sales stopped until modifications are approved.
Automakers must submit complete inspection and testing reports and integrate them into production consistency systems. Additionally, companies must establish accident-reporting mechanisms that require incidents to be reported within 24–48 hours.
It further addresses that OTA updates affecting essential functions must be registered and approved; emergency updates require suspending sales until corrective measures are approved; and a sandbox system monitors safety risks.
This structure makes OTA releases a regulated product. Software teams must build in time for pre‑authorization and maintain detailed documentation for each update.
Impact of Guidelines on ADAS Development
Marketing versus product reality
For years, marketing departments pitched near‑autonomous capabilities while engineering teams were still tackling edge cases. The new directives force alignment between messaging and capability. Companies cannot imply autonomy; they must state the exact level of assistance, emphasise driver responsibility, and provide training at handover.
The draft standard requires confirmation that drivers complete training each time the system is activated. This may slow adoption in the short term, but it reduces confusion and risk. For international executives, the message is clear: treat compliance messaging as a core product feature.
Engineering accountability and safety cases
The requirement for driver‑monitoring with hands‑off detection and fallback transitions pushes engineering teams to design systems with a safety case approach similar to aviation or rail. Systems must degrade gracefully, issue warnings, and shut down or lock out when misuse persists.
Data logging becomes essential not only for performance improvement but also for forensic analysis.
MIIT’s rules effectively demand an end‑to‑end evidence chain from simulation through track tests to real‑world logs. Companies need to maintain a traceable dossier for each function release, covering scenario coverage, failure mode analysis, and rollback plans. This raises costs but also builds trust with regulators and the public.
Supply‑chain and hardware implications
Functions now subject to approval include not only software but also hardware changes. Upgrades that alter major technical parameters, such as sensor types or configurations, require regulatory approval.
Unapproved changes can trigger recall procedures. Thus, procurement and product planning must stabilise sensor and compute configurations over longer cycles to avoid repeated approval loops. The hardware lock also encourages modular, upgradable architectures, allowing new sensors to be integrated with minimal changes to safety‑critical behaviours.
How is BYD Adopting the Regulation
BYD has aggressively embraced the change. In February 2025, the company announced it would equip 21 models with its God’s Eye driver‑assistance suite, offering the system across the entire lineup at no extra cost.
BYD’s chairman, Wang Chuanfu, said that “good technology should be available to everyone” and predicted high‑level smart driving would become as essential as seatbelts and airbags within two to three years.
The system is tiered: God’s Eye C (DiPilot 100) uses 12 cameras, five millimetre‑wave radars, and 12 ultrasonic sensors; God’s Eye B adds LiDAR; and God’s Eye A employs three LiDAR sensors with the powerful DiPilot 600 computing system. Prices start as low as CNY 69 800 (about USD 9 550) for models like the Seagull.
Investment in engineering and data
BYD backs its ambitions with massive resources. The company employs 110,000 engineers, of whom 5,000 are dedicated to intelligent driving R&D. Their system processes data from 72 million kilometres of daily training mileage.
Such a scale gives BYD an advantage when building scenario libraries for regulatory evidence. Many Chinese start‑ups rely on small fleets and simulation; BYD uses both to train and verify its models, with millions of kilometres of real‑world operation.
Bundling OTAs and standardising hardware
In July 2025, BYD announced what it called the “largest‑scale smart driving OTA in history”. The update covered more than one million vehicles already equipped with God’s Eye.
BYD emphasised that bundling functions into a single major release would make the system safer and smarter, aligning with MIIT’s requirement that OTA updates affecting driving must be pre‑authorized.
The company operates three hardware tiers, which simplify validation across its lineup. It also deepened its LiDAR supply relationship with Hesai, announcing that more than ten models in 2025 would carry Hesai units. Standardised hardware reduces the number of approval permutations when regulators review changes.
Accepting liability for bounded domains
Perhaps the most striking response was BYD’s decision in July 2025 to assume responsibility for any accidents or losses arising from its smart parking feature. The company stated that users would not need to go through insurance; BYD would handle claims directly.
The smart parking function is described as having Level 4 capabilities in confined areas, allowing the driver to take their hands and eyes off the wheel under certain conditions. By assuming liability, BYD signals confidence in its technology and seeks to alleviate regulator concerns about unsupervised features.
This risk‑transfer strategy mirrors Mercedes‑Benz’s approach with its Drive Pilot Level 3 system in Germany, where the manufacturer accepts responsibility when the system is in control. For other companies, the lesson is that carefully bounded features with manufacturer liability can earn regulators’ trust.
Balancing innovation and caution
Chinese regulators have indicated that they still support innovation. A Reuters report noted that officials aim to resume Level 3 validation tests in 2026 and are working with companies such as Huawei and Changan to draft safety rules for higher levels.
The same report noted that more than 60% of new cars sold in China in early 2025 had Level 2 features. BYD’s approach—scaling Level 2 features while building the groundwork for Level 3—aligns with this timeline.
The company rolled out God’s Eye across its range for free, creating a massive fleet of vehicles that can collect data and improve algorithms. With regulators limiting public beta programmes, BYD’s in‑house data becomes even more valuable for closed‑loop improvement.
Conclusion
China’s 2025 directive ended the free‑for‑all era of assisted‑driving marketing and demanded proof instead of promises. By banning inflated terminology, restricting public betas and unsupervised features, mandating hands‑on detection, and regulating software updates, MIIT reset the industry’s priorities.
The changes align with global movements in Europe and the United States, where regulations like UNECE DCAS, Euro NCAP’s 2026 protocols, and NHTSA’s AEB rule push for robust driver monitoring, low‑light performance, and evidence‑based safety.
For ADAS management teams, the reset is an opportunity to build better products. BYD’s response shows that companies can adapt quickly.
Others must follow by aligning language and HMI, building regulator‑ready evidence kits, stabilising hardware, integrating driver‑monitoring from the start, and engaging regulators and customers.
Looking ahead, the next three years will likely see China approve the first Level 3 systems under strict supervision, global convergence on driver‑monitoring and dark‑scenario performance, consolidation among suppliers, near‑universal adoption of ADAS, and new data‑governance rules.
Those who invest in evidence and trust will not only comply with China’s hard rules but also gain a strategic advantage worldwide. In the end, the shift from hype to hard rules might be the best thing that ever happened to assisted driving.
How Can We Help You?
We support industry-leading R&D and Innovation professionals through complex problems. Describe your challenge, and let us bring clarity and expertise.
