Wang Tiger, architect of Baidu Automotive Security Solution, introduced that Baidu Intelligent Cloud has achieved a continuous layout in the field of artificial intelligence, and the cloud-intelligent integration 3.0 architecture has achieved a full-stack layout on the four layers of technology stacks: chip layer, framework layer, model layer and application layer.
Source: Speaker material
Mr. Wang Tiger introduced a lot of thinking and practice in the field of vehicle network security based on the accumulation of AI capabilities over the years. The TARA analysis platform based on large models and the automated and intelligent Internet of Vehicles security operation platform (VSOC) are introduced. He pointed out that compared with the cumbersome and complex traditional manual TARA analysis, the TARA analysis platform based on large model reconstruction can more efficiently and comprehensively realize the TARA analysis needs of enterprises. Taking the Internet of Vehicles Security Operation Platform (VSOC) as an example, it shows that large models can help car companies greatly improve the efficiency of security operations, so that security operations can shift from compliance to automated and intelligent operations.
Wang Tiger | Baidu Automotive Safety Solution Architect
The following is a summary of the speech:
Baidu security team overview
As a professional security department under Baidu Group, Baidu Security Team focuses on providing comprehensive security solutions for Baidu and its partners. The team’s business covers multiple fields such as basic security, data security, business security, and Internet of Vehicles security, forming a complete product matrix. In the field of Internet of Vehicles security, Baidu’s security team is committed to improving the security protection capabilities of cars through technological innovation and ensuring the security and privacy of user data.
Source: Speaker material
Source: Speaker material
Compliance rigid need: security upgrade under the triple drive
Global regulations are strengthening the safety responsibilities of car companies throughout the chain. The EU WP29 R155 regulation requires that models that have not passed the network security certification are prohibited from accessing the Internet of Vehicles, and the domestic “GB 44995” mandatory standard clearly requires the establishment of a threat monitoring and risk control system. At the market level, China, January-May 2025New energy vehiclessales account for more than 60% of the total market, and its intelligent characteristics promote the deep integration of safety technology.
TARA Analytics Revolution: AI Reimagines Risk Assessment Processes
As a knowledge-enhanced AI technology base, Baidu Wenxin model brings innovation to the Internet of Vehicles security system. The Internet of Vehicles security platform system built by Baidu takes “multiple systems and one platform” as the core, integrates full-cycle security management, personnel system construction and compliance control capabilities, and realizes key functions such as security design and implementation and penetration testing based on Baidu’s AI base.
Among them, the reconstruction of TARA analysis methods by large models is particularly critical, and traditional manual TARA analysis faces four major bottlenecks: cumbersome process and high analysis complexity, lack of depth and comprehensiveness, consistency and standardization challenges, and high maintenance costs for knowledge updates. Baidu automatically identifies vehicle asset information through AI, and large models automatically assess potential hazards and threat scenarios, generate risk handling strategies and security objectives, significantly improving analysis efficiency and accuracy. This technological breakthrough not only solves the problem of human dependence on traditional methods, but also realizes the accurate derivation of security requirements through intelligent means, providing more efficient technical support for the security protection of the Internet of Vehicles.
Source: Speaker material
Baidu launched a dual-engine architecture based on the Wenxin X1 inference model: the fast thinking system calls the pre-trained TARA analysis security knowledge base through RAG technology to complete the matching of known threats within 5 minutes; The slow thinking system uses the Know-How knowledge base to conduct in-depth thinking and analysis of large models in strict accordance with the seven-step analysis methodology defined by the ISO/SAE 21434 standard for new assets. This solution can achieve fully automated analysis of large models for more than 90% of tasks, and the measured efficiency is increased by more than 10 times.
Source: Speaker material
VSOC Security Operations: AI for combat-level defense
In order to address the challenges of automotive compliance, Baidu’s security team has launched a compliance-oriented automotive cybersecurity operation platform. The platform has gone through multiple stages of development, from meeting compliance requirements to implementing security attack and defense, and then to intelligent automated security operations supported by AI. At present, the platform can not only meet the compliance requirements of automotive network security, but also provide efficient and intelligent security operation services for automakers.
Source: Speaker material
The Internet of Vehicles security operation platform needs to go through three stages of evolution: compliance detection, offensive and defensive practice, and AI automation. Traditional VSOC faces multiple challenges: heavy manual processing workload, serious alarm errors and noise, high personnel safety capabilities, and poor alarm handling flexibility.
Source: Speaker material
Baidu’s AI solution improves operational efficiency through three levels of capabilities: the AI research and judgment engine automatically analyzes and judges the security alarms reported by IDPS based on large models to identify whether there are real security risks and risk levels; The alarm noise reduction module is based on the capabilities of AI large models to realize noise reduction processing such as whitening, deduplication, aggregation, and automatic ignoring of alarms. Copilot supports natural language commands and automates functions such as risk statistics, report generation, risk interpretation, work order creation, and security response.
Source: Speaker material
Future challenges and industry evolution
The dynamic offensive and defensive environment continues to improve technical requirements, and AI still needs manual assistance in unstructured document parsing. After the implementation of the new regulations in 2026, in order to reduce the cost of compliance construction and improve the efficiency of compliance operations, AI-driven compliance capabilities will become an essential element of the compliance development and overseas process of car companies.
(The above content comes from the keynote speech “Exploration and Thinking of Large Models in Automotive Compliance” delivered by Wang Tiger, architect of Baidu’s automotive security solutions, at the 4th China Internet of Vehicles Security Conference in 2025 on June 19, 2025.) ）
