Written by Pavan Vemuri- Director of Product Engineering
Safe Thinking: How to Govern AI in Tomorrow’s Vehicle Systems
The automotive industry stands at a critical crossroads as artificial intelligence transforms how vehicles operate and interact with humans.
As manufacturers integrate increasingly sophisticated AI systems, from advanced driver assistance features to emotionally responsive interfaces, the sector faces complex challenges in ensuring these technologies are safe and ethically governed. This article lists the critical safety considerations, a framework for responsible implementation, and implementation best
practices to be considered when implementing AI in vehicles while maintaining
appropriate safety standards and governance frameworks.
The Evolving AI Landscape in Automotive Systems
Modern vehicles have evolved from purely mechanical systems to complex computing platforms. Today’s cars contain up to 100 million lines of code and numerous AI applications, ranging from practical to experiential. The following capabilities illustrate how comprehensively AI has redefined the vehicle experience:
- Advanced driver-assistance systems (ADAS) that monitor surroundings and assist with driving tasks
- Predictive maintenance systems that analyze vehicle performance data
- Personalized comfort systems that adjust vehicle settings based on driver preferences
- Voice assistants and natural language processing for hands-free control
- Emotional and behavioral monitoring systems that assess driver state
- AI transforming aftersales operations across OEMs
The latest frontier involves AI systems that aim to create more emotionally-aware experiences. LG’s “Affectionate Intelligence” introduced at CES 2025 represents this trend, with features that monitor driver health, suggest schedule adjustments, and create customized in-vehicle experiences based on detected preferences and needs. Similarly, companies like BMW and Mercedes-Benz are developing systems that recognize driver emotions and respond with appropriate environmental adjustments. Realizing that promise, however, demands confronting a set of challenges that are as consequential as the technology itself.
Critical Safety Considerations for Automotive AI
Functional Safety vs. Safety of the Intended Functionality
When implementing AI in vehicles, manufacturers must address two distinct safety
domains:
• Functional Safety (ISO 26262): Ensuring systems operate correctly when
components fail
• Safety of the Intended Functionality (ISO 21448/SOTIF): Addressing risks that occur even when systems work as designed
The SOTIF standard is particularly relevant for AI systems, as it addresses hazards
caused by inherent limitations rather than component failures. According to ISO, SOTIF
covers “the absence of unreasonable risk due to hazards resulting from functional insufficiencies of the intended functionality or by reasonably foreseeable misuse by persons.”
This distinction becomes crucial for AI systems that make complex decisions based on
sensor inputs and algorithmic processing. A system might function exactly as programmed yet still create safety hazards if its decision-making processes don’t account for all possible scenarios.
Challenges faced by AI applications within the Automotive Industry
Safety Challenges
• Performance Limitations in Diverse Conditions: AI systems trained on limited datasets may perform poorly in conditions outside their training parameters. This is particularly concerning for systems that interpret visual information, as lighting conditions, weather, and unusual objects can all impact performance.
• An American Automobile Association (AAA) study found that automatic emergency
braking systems with pedestrian detection were “completely ineffective at night,” highlighting how environmental factors can impact AI performance.
• Human, Machine Interaction and Distraction: As vehicles incorporate more AI driven interaction features, they risk creating new forms of driver distraction. Systems that actively engage with drivers, suggesting destinations, offering schedule adjustments, or responding to emotional states, must be designed to minimize distraction while still providing value.
Transparency and Predictability: Effective human machine cooperation requires
that AI systems operate in ways drivers can understand and predict. When an AI system makes a decision, whether adjusting vehicle dynamics, suggesting a route change, or modifying cabin conditions, the driver should understand why this is happening and be able to anticipate similar actions in the future. If drivers cannot understand why systems are responding in certain ways, they may experience confusion or frustration that undermine the intended benefits.
Governance Challenges
Data Privacy and Protection: Modern vehicles are “data centers on wheels”
collecting enormous amounts of information about drivers, passengers and other aspects. A 2023 investigation by the Mozilla Foundation found some concerning privacy practices across the automotive industry, with all 25 car brands they reviewed receiving their “Privacy Not Included” warning label. The study found that 84% of car brands shared personal data with service providers and data brokers, and 76% admitted to selling customer data. This situation is especially problematic for emotionally-aware AI systems, which may collect particularly sensitive data about driver states, preferences, and behaviors. Without strong privacy protections, this information could be exploited for commercial purposes or create security vulnerabilities.
Regulatory Compliance Across Jurisdictions: Just like web applications and mobile applications, Auto manufacturers integrating AI systems into the vehicular experience must follow different privacy and data rules that are applicable to each major region. Europe has the strictest rules with GDPR and the upcoming AI Act that labels vehicle AI as “high risk.” The United States mainly regulates at the state level, with California having the strongest protections and the FTC paying more attention to car data practices. China requires all data collected from cars in China to be stored inside Chinese borders. Car makers need to build flexible systems that work everywhere while still following these rules.
Ethical Use of AI and Algorithmic Fairness: AI applications within vehicles should be avoiding built-in biases and should be fair and ethical with their outcomes. Drivers should always remain in control, not the AI. Companies must think about how their AI suggestions might change driver behavior. This is especially important for systems that respond to emotions, which should never manipulate drivers or encourage unsafe habits based on how they’re feeling.
A Framework for Responsible Implementation
1. Safety first Design Approach: Lay heavy emphasis on safety methodologies throughout development. Implement comprehensive testing across diverse operational conditions. Design for graceful degradation when systems face uncertainty.
Example: BMW’s approach to developing driver monitoring systems demonstrates this principle. Their Driver Attention Camera system uses AI to detect driver distraction but is designed to degrade gracefully when faced with unusual conditions like the driver wearing sunglasses, ensuring safety is maintained even when optimal detection isn’t possible.
2. Privacy-Preserving Architecture: Process sensitive data locally wherever possible. Implement principles which enable to only collect necessary information and be as minimalistic as possible. Provide clear, accessible instructions and controls for users to manage data collection.
Example: Tesla has implemented local AI processing for certain vision-based features, allowing the vehicle to make decisions without sending sensitive visual data to external servers.
3. Transparent Governance: Clearly document system capabilities, limitations,
and data practices. Establish internal ethics committees to ensure ethics are integrated into feature development. Participate in industry standardization approaches.
Example: Volvo’s approach to transparency offers a positive example. The company provides detailed documentation of their driver assistance systems, explicitly communicating limitations and expected behaviors to help drivers understand what these systems can and cannot do.
Implementation Best Practices
1. Rigorous testing and validation:
Implement comprehensive testing regimes that include:
a. Simulation based testing across thousands of scenarios.
b. Physical testing in controlled environments
c. Limited real-world testing with careful monitoring.
2. Clear User communication
a. Inculcate transparency in explaining system capabilities and limitations.
b. Use intuitive interfaces that make AI decision making transparent.
c. Implement progressive disclosure that introduces complex features gradually.
3. Continuous Monitoring and Improvement
a. Establish feedback channels for users to report issues.
b. Implement telemetry systems that identify potential problems.
c. Create rapid response capabilities for addressing emergent issues.
The Road Ahead
As AI in vehicles continues to evolve from driver assistance to emotional awareness, the
industry faces significant challenges in ensuring these systems enhance the driving
experience while respecting safety, privacy, and ethical considerations. By establishing
strong governance frameworks now, manufacturers can build trust with consumers and
regulators while pushing technological boundaries.
The potential benefits of advanced automotive AI are substantial, from improved safety
through better situational awareness to more comfortable, personalized experiences.
Realizing these benefits will require thoughtful collaboration between technology
developers, automotive manufacturers, regulators, and end users.
With appropriate safety measures and governance frameworks, automotive AI can
enhance the driving experience without compromising safety or privacy. The key lies in
developing systems that serve as effective assistants that understand not just what
drivers need, but how to deliver it in a way that respects their autonomy, safety, and
personal boundaries.
References
American Automobile Association. (2022). “Automatic Emergency Braking with
Pedestrian Detection.” AAA Research.
International Organization for Standardization. (2022). “ISO 21448:2022 – Road vehicles- Safety of the intended functionality.” https://www.iso.org/standard/77490.html
LG Electronics. (2025, January 6). “LG Unveils a Day in a Life with ‘Affectionate Intelligence’
LG World Premiere.” LG Newsroom. https://www.lgnewsroom.com/2025/01/lg-unveils-a-day-in-a-life-with-affectionate-intelligence-at-lg-world-premiere/
Mozilla Foundation. (2023). “It’s Official: Cars Are Terrible at Privacy and Security.” Privacy Not Included. https://foundation.mozilla.org/en/privacynotincluded/articles/its-official-cars-are-the-worst-product-category-we-have-ever-reviewed-for-privacy/
Mercedes-Benz – 2026 in-car AI (MBUX) https://group.mercedes-benz.com/technology/innovation/collaboration/liquid-ai.html
Tesla AI
https://www.tesla.com/AI
Interested in exterior vehicle sensing technology?
With a pass to InCabin USA, you’ll also get full access to our
co-located sister event, AutoSens. Find out more here >>
