The world automotive industry is changing rapidly: switching to automation, connectivity, new software and more complex vehicle architectures. Automation is transforming cars into smarter connected systems that can operate as information processing centres, react according to circumstances and support the driver. This rising complexity brings increasingly significant safety challenges, as we explored at the 2026 TEDA Forum, where industry specialists took a close look at soaring recall volumes, software issues, battery and electric-drive problems and evolving regulation.

The changing recall landscape also shows how vehicle safety is moving beyond traditional mechanical failures. China recorded 121 million cumulative vehicle recalls by the end of 2025, while regulatory defect investigations accounted for 53% of those recalls. At the same time, thermal safety has improved significantly, with the fire rate per 10,000 new energy vehicles falling by 68% between 2019 and 2025. As mechanical problems decline, software, battery management, sensors, driver-assistance systems, cybersecurity, and other connected technologies are becoming increasingly important areas of safety oversight.

1. Rising Vehicle Recalls and the Changing Safety Landscape

At the 2026 TEDA Forum on September 19, Xiao Lingyun, head of the industrial products institute at the State Administration for Market Regulation’s Defective Product Recall Technology Center, presented important data on China’s recall situation. By the end of 2025, China had recorded 121 million vehicle recalls cumulatively, with 53% associated with regulatory defect investigations. With approximately 370 million vehicles on Chinese roads by June 2026, the scale of recalls demonstrates how widespread vehicle safety monitoring has become. The changing nature of vehicle technology is also making the safety environment more complex.

Key Recall Trends:

• 121 million cumulative recalls
• 53% linked to defect investigations
• 370 million vehicles by June 2026
• 68% decline in NEV fire rate
• Growing software-related defects

Thermal safety provides an important example of progress within this changing environment. Between 2019 and 2025, the fire rate per 10,000 new energy vehicles in China declined by 68%, suggesting that targeted thermal regulations and hardware standards have produced significant improvements. However, the decline in traditional mechanical defects does not eliminate safety risks. Instead, recall problems are increasingly connected to software bugs, battery-management systems, and driver-assistance technologies, creating a different type of safety challenge for manufacturers and regulators.

2. Software and Intelligent Driving Become Major Recall Sources

The growing importance of software was illustrated dramatically on August 21, 2026, when nearly 8 million vehicles in the Chinese market were subject to recalls in a single day. Many of the issues were associated with software glitches, battery-system problems, and intelligent-driving anomalies rather than conventional mechanical failures. This shift demonstrates how the increasing amount of digital technology inside modern vehicles can create new categories of defects that traditional automotive quality-control systems were not originally designed to manage.

Modern Recall Challenges:

• Software glitches
• Battery system flaws
• Intelligent-driving anomalies
• Driver-assistance defects
• Complex vehicle electronics

The expansion of intelligent systems means that a vehicle’s safety increasingly depends on interactions between software, sensors, hardware, communications, and decision-making systems. A defect in one component can potentially affect the behavior of several connected functions. SAMR data showed that approximately 2.56 million vehicles in China were recalled in 2024 because of driver-assistance system issues, representing 23% of total recalls. This growing contribution from software-based systems highlights why conventional inspection and quality-control approaches need to evolve alongside vehicle technology.

3. Human Intelligence and the Challenge of Vehicle Intelligence

Understanding the limitations of intelligent vehicles can be approached by examining how intelligence itself has been studied. In the early 1900s, psychologist Charles Spearman observed that people who performed well in one mental task often performed well in others, leading him to propose a general intelligence factor known as “g.” This work contributed to the development of modern IQ testing, in which scores follow a bell curve. The source notes that scores between 85 and 115 represent average cognitive performance for about 82% of the population, while scores of 130 or above represent gifted status for roughly 2%.

Human Intelligence Concepts:

• Charles Spearman’s general intelligence
• General intelligence factor “g”
• Modern IQ testing foundations
• Average range of 85-115
• Gifted threshold of 130 or above

In the 1940s, psychologist Raymond Cattell further divided intelligence into fluid and crystallized forms. Fluid intelligence involves reasoning through unfamiliar problems without relying heavily on previously acquired knowledge, while crystallized intelligence represents accumulated facts, knowledge, and skills. According to the source, fluid intelligence tends to peak in early adulthood and decline afterward, whereas crystallized intelligence can continue developing throughout life. These distinctions provide a useful framework for considering why automated systems may perform well in familiar circumstances but struggle when they encounter unexpected situations.

4. Autonomous Vehicles and the Limits of Standardized Learning

When cognitive theories are applied to autonomous and driver-assistance systems, an important technical challenge becomes visible. Automated vehicles depend heavily on knowledge gathered through training data and testing under known or relatively standard driving conditions. This resembles crystallized knowledge because the system relies on previously collected examples and learned patterns. However, unusual hazards may require something closer to fluid reasoning, where a system must respond effectively to circumstances that do not closely match its previous experience.

Autonomous-System Challenges:

• Dependence on training data
• Standard driving conditions
• Unexpected road hazards
• Irregular objects
• Limited adaptive reasoning

The automotive industry previously assumed that many driver-assistance challenges could be addressed through existing systems, but real-world experience has demonstrated that vehicles can encounter difficulties when confronted with irregular objects and unusual situations. The distinction between learned information and flexible reasoning therefore becomes increasingly important. Advanced driver-assistance systems must interpret complex environments, make decisions, and execute responses in real time. Developing algorithms capable of handling these unexpected situations remains one of the central challenges associated with intelligent vehicle development.

5. Multiple Forms of Intelligence and Metacognition

Other cognitive theories also provide different ways of considering operational intelligence. Howard Gardner identified multiple forms of intelligence, including verbal-linguistic, logical-mathematical, spatial-visual, bodily-kinesthetic, musical, interpersonal, intrapersonal, and naturalist abilities, with a ninth existential dimension also discussed. Robert Sternberg proposed a triarchic framework in which real-world intelligence involves balancing analytical, creative, and practical thinking. These theories illustrate that intelligence is not necessarily a single capability but can involve several different forms of processing and adaptation.

Cognitive Frameworks:

• Gardner’s multiple intelligences
• Spatial and logical abilities
• Interpersonal and intrapersonal skills
• Sternberg’s triarchic framework
• Analytical, creative, practical thinking

Daniel Goleman’s model adds emotional and social capacities, including self-awareness, self-regulation, motivation, empathy, and social skills. The source notes research suggesting that emotional and social skills can be four times more predictive of career achievement than IQ alone. Cultural perspectives on intelligence also differ, with American views emphasizing areas such as verbal fluency and planning, Confucian and Taoist traditions emphasizing politeness, discipline, and social harmony, and the Luo people of East Africa combining cleverness with practical thinking, social responsibility, and instruction comprehension. These differences reinforce the complexity of defining intelligence across contexts.

6. Metacognition and the Need for Adaptive Vehicle Systems

Higher cognitive functioning also depends on metacognition, which refers to the ability to monitor and evaluate one’s own thought processes. Human thinking involves automatic background processes as well as deliberate reflection, and metacognition allows people to identify mistakes and adjust their strategies. Research published in Philosophical Transactions of the Royal Society B has demonstrated that deliberate reflection can improve group decision-making by encouraging people to explain and evaluate their reasoning. A similar capability remains difficult to reproduce within autonomous vehicle software.

Metacognition Challenges:

• Monitoring thought processes
• Identifying errors
• Evaluating decisions
• Adapting strategies
• Monitoring perception accuracy

For autonomous vehicles, algorithmic metacognition could mean enabling software to evaluate how accurately its perception systems are interpreting the surrounding environment. A vehicle might need to recognize uncertainty in its sensor information and respond appropriately rather than treating every interpretation as equally reliable. Developing such monitoring capabilities remains a major engineering challenge. The source also connects this issue with the Flynn Effect, the 20th-century trend of rising IQ scores associated with improvements in healthcare, nutrition, and education. As environmental improvements plateaued in advanced nations, some regions experienced stalled or reversed gains while developing regions continued to show increases, illustrating how intelligence can be influenced by surrounding conditions.

7. Legacy Recall Rules Face New Automotive Technologies

Traditional recall regulations were developed largely during the fossil-fuel era and relied on laws such as the Product Quality Law and conventional defective-vehicle recall procedures. The traditional process generally follows a linear sequence in which an accident or consumer complaint occurs, regulators investigate the issue, and a recall is eventually initiated. While this framework provides an established safety mechanism, its reactive nature can create delays between the emergence of a problem and regulatory intervention, particularly when vehicles are operating in large numbers on public roads.

Legacy Regulatory Challenges:

• Traditional defect investigations
• Accident-driven reporting
• Consumer complaints
• Delayed regulatory intervention
• Fossil-fuel-era frameworks

Connected and intelligent vehicles create additional problems that are not easily covered by older regulations. As battery swapping expands, for example, existing rules may focus on the complete vehicle manufacturer without clearly defining responsibility for batteries that can be exchanged between vehicles. Autonomous-driving software may also be concentrated among specialized suppliers, potentially limiting automakers’ visibility into internal software code and creating questions about supplier responsibility. Cybersecurity, data privacy, and wireless network risks introduce further concerns that were largely absent from conventional vehicles.

8. Sandbox Testing and the Three-Line Safety Framework

To modernize oversight, the State Administration for Market Regulation and the Ministry of Industry and Information Technology introduced accident reporting and sandbox supervision. Xiao Lingyun described sandbox monitoring as an important development for smart-car safety because it provides a controlled, high-stress environment outside normal public traffic. Vehicles can be subjected to difficult conditions designed to push their systems toward technical limits, allowing regulators and manufacturers to identify potential problems before they become real-world safety incidents.

Sandbox Supervision Benefits:

• Controlled testing environment
• High-stress vehicle evaluation
• Early defect detection
• Software and sensor testing
• Approximately 200 vehicles tested

Around 200 vehicles have undergone sandbox testing, allowing regulators to identify software and sensor defects before market launch. Automakers then used the findings to improve their products proactively, helping address potential problems before affected vehicles entered wider operation. Mainstream manufacturers are increasingly enrolling new intelligent models in the program. Xiao Lingyun’s “three-line theory” places sandbox testing at the high line, where risks can be detected before accidents; daily accident reporting at the middle line; and recalls at the bottom line when problems eventually surface. This creates a more layered approach to safety monitoring.

9. AI Data Loops, National Standards, and Global Recall Activity

Regulators are increasingly using artificial intelligence and connected-vehicle data to strengthen safety monitoring. SAMR engineers integrated accident reporting with a hazard-scenario database, allowing AI tools to automatically generate testing scenarios based on historical crash data. Seven or eight automakers are evaluating these generated scenarios within sandbox environments, helping accelerate the identification of risks associated with Level 2 and Level 3 driver-assistance systems. At the same time, regulators are developing national standards covering defect identification for L2 and L3 systems and standardized accident-reporting data formats.

Advanced Safety Oversight:

• AI-generated test scenarios
• Connected-vehicle data loops
• L2 and L3 safety standards
• Standardized accident reporting
• Battery quality standards

SAMR also collaborated with CATL to release four national standards for power-battery quality management. These standards aim to formalize leading corporate practices into national requirements while supporting their further development toward international rulemaking. Similar recall activity is occurring in other markets. Recent NHTSA filings in the United States include recalls involving General Motors, Ford, BMW, Mercedes-Benz, Toyota, Tesla, Volkswagen, Volvo, Audi, and other manufacturers. The reported issues range from software and display problems to mechanical, structural, battery, braking, lighting, seat-belt, and component defects.

10. International Compliance and the Future of Intelligent Vehicle Safety

Recall activity is becoming increasingly important as Chinese vehicle exports expand globally. China Association of Automobile Manufacturers data cited in the source shows that China exported 1.758 million new energy vehicles from January through September, representing an 89.4% year-on-year increase. Zhang Hong of the China Automobile Dealers Association explained that complex driver-assistance systems combine perception, decision-making, and execution technologies, creating opportunities for unexpected failures and software-hardware compatibility problems. He also noted that identifying safety issues through recalls can encourage automakers to increase research and development investment and accelerate technological improvements.

Global Safety Priorities:

• Expanding NEV exports
• International recall compliance
• Driver-assistance reliability
• Overseas regulatory requirements
• Consumer confidence

The Defective Product Recall Management Center created an export compliance platform to consolidate overseas recall requirements and help manufacturers address foreign regulations earlier in vehicle development. This is increasingly important as global price competition since 2026 pushes automakers to launch new models faster without reducing safety attention. As vehicles become more intelligent and connected, safety oversight must cover traditional hardware as well as software, sensors, batteries, driver assistance, cybersecurity, and hardware-software interactions. Testing sandboxes, accident reporting, AI-generated scenarios, national standards, and international compliance tools are becoming important ways to ensure safety remains central to intelligent vehicle development.