Autonomous Vehicle AI Liability

The Autonomous Vehicle Liability Reckoning

#

Autonomous vehicle technology promised to eliminate human error, responsible for over 90% of crashes. Instead, a new category of liability has emerged: algorithmic negligence, where AI systems make fatal errors that cannot be easily explained, predicted, or prevented. As self-driving technology scales from test fleets to consumer vehicles, courts are grappling with fundamental questions: Who bears responsibility when software kills? What disclosure duties exist for AI limitations? And does the promise of autonomy shift liability from driver to manufacturer?

The August 2025 $329 million jury verdict against Tesla, the first to find Autopilot defective in a wrongful death case, signals that the era of manufacturer accountability for AI driving systems has arrived.

The Landmark Tesla Verdict: Benavides v. Tesla

#

The Facts

#

On April 27, 2019, George Brian McGee was driving his Tesla Model S on Card Sound Road in Key Largo, Florida, with Autopilot engaged. After dropping his phone and looking away from the road, the vehicle struck 22-year-old pedestrian Naibel Benavides Leon, killing her, and severely injured her boyfriend Dillon Angulo.

The case went to trial in federal court in Miami in July 2025, the first wrongful death lawsuit involving Tesla’s driver-assistance systems to reach a jury.

The Verdict

#

On August 1, 2025, the jury found Tesla partially liable and awarded the plaintiffs $329 million in total damages:

• $129 million in compensatory damages
• $200 million in punitive damages

The jury assigned 33% of fault to Tesla and 67% to driver McGee, who admitted taking his eyes off the road. Tesla’s share of the damages totals approximately $243 million.

Historic Significance

#

“This is the first time that Tesla has been hit with a judgment in one of the many, many fatalities that have happened as a result of its Autopilot technology,” noted Alex Lemann, a professor at Marquette University Law School.

U.S. District Judge Beth Bloom, in allowing the case to proceed to trial, wrote that “a reasonable jury could find that Tesla acted in reckless disregard of human life for the sake of developing their product and maximizing profit.”

The Plaintiffs’ Theory

#

The plaintiffs successfully argued three key defects:

1. Unsafe road activation: Autopilot could be engaged on roads where it shouldn’t operate safely, including unlit rural highways without clear lane markings

2. Inadequate driver monitoring: The system failed to adequately detect and respond to driver inattention

3. Marketing misrepresentation: CEO Elon Musk oversold Autopilot’s capabilities as safer than it actually was, creating dangerous overreliance

Tesla’s Response and Appeal

#

Tesla denounced the verdict and filed for appeal, stating: “Today’s verdict is wrong and only works to set back automotive safety and jeopardize Tesla’s and the entire industry’s efforts to develop and implement life-saving technology.”

Tesla is seeking to have the verdict tossed or obtain a new trial, citing “substantial errors of law and irregularities at trial.”

Tesla Autopilot/FSD: Standard of Care Analysis

#

The Tesla litigation landscape provides a critical case study in AI driver-assistance standard of care, illuminating the gap between marketing promises and technical reality, the regulatory response to that gap, and the emerging professional standards that automakers deploying AI driving systems must meet.

Marketing vs. Reality: The Fundamental Problem

#

Tesla markets two driver-assistance features:Autopilot (included on all vehicles) and Full Self-Driving (FSD) (a $12,000-$15,000 upgrade), using terminology that implies capabilities these systems do not possess.

The Classification Gap:

As UC Berkeley Professor Scott Moura explained: “Tesla has a technology product that they have branded as FSD, which is ‘fully self-driving.’ There are levels of automated driving. One-two-three-four-five. But their FSD tech corresponds to Level Two, not Level Five. Thus, one can argue that it is misleading.”

The Disclosure Problem:

Tesla’s disclaimers state that “active human supervision” is required. But the company simultaneously:

• Uses terms like “Autopilot” and “Full Self-Driving” in marketing
• Released videos showing hands-free driving
• CEO Elon Musk repeatedly promised imminent full autonomy (since 2016)

This creates what the Benavides jury found: dangerous overreliance by drivers who believe their vehicles can do more than they actually can.

NHTSA Investigations: Scale of the Problem

#

Tesla’s driver-assistance systems have been the subject of multiple federal investigations, revealing patterns of failures that inform standard of care analysis.

Current NHTSA Investigations (2025):

NHTSA’s Design Critique:

NHTSA reviewed 956 crashes where Autopilot was alleged to be in use (446 detailed, 510 supplemental). The agency’s conclusion was damning:

“A comparison of Tesla’s design choices to those of L2 peers identified Tesla as an industry outlier in its approach to L2 technology by mismatching a weak driver engagement system with Autopilot’s permissive operating capabilities.”

In other words: Tesla allowed Autopilot to operate in more situations than competitors while providing less robust driver monitoring, a design choice that violates emerging standard of care.

Fatality Statistics:

As of October 2025:

• 65 reported fatalities linked to Autopilot or FSD
• 54 of those fatalities are under active NHTSA investigation
• Hundreds of nonfatal incidents documented

DOJ Criminal Investigation

#

Beyond civil and regulatory liability, Tesla faces potential criminal exposure for its marketing claims.

Investigation Scope:

The Department of Justice has been investigating Tesla since 2021, with expanded subpoenas issued in 2023. According to Reuters, prosecutors are examining whether Tesla committed:

• Wire fraud: Deception in interstate communications regarding self-driving capabilities
• Securities fraud: Material misstatements to investors about technology readiness

Key Issues Under Review:

• Statements by Tesla and CEO Elon Musk suggesting vehicles can drive themselves
• Marketing materials implying higher autonomy than systems actually provide
• Timing and accuracy of safety disclosures

Current Status:

Tesla disclosed in SEC filings that to its knowledge, “no government agency has concluded that any wrongdoing happened in any ongoing investigation.” However, prosecutors reportedly continue evidence-gathering as of late 2025.

California DMV Deceptive Marketing Case

#

California’s Department of Motor Vehicles has pursued the most aggressive regulatory action against Tesla’s marketing practices.

Case Timeline:

DMV’s Core Arguments:

1. The terms “Autopilot” and “Full Self-Driving Capability” imply full autonomy
2. Drivers must remain “fully engaged and attentive at all times”, contradicting the marketing impression
3. This violates California Vehicle Code provisions banning misleading marketing of partially automated features

Tesla’s Defenses:

• Disclaimers require “active human supervision”
• “Self-driving” is “aspirational” rather than deceptive
• The state allowed Tesla to use these terms for years, granting “implicit” approval
• Marketing statements are protected free speech under the First Amendment

Potential Consequences:

If the DMV prevails, Tesla could face a 30-day suspension of manufacturing operations at its Fremont, California factory, a significant financial and reputational blow.

International Enforcement

#

Tesla faces parallel enforcement actions globally:

France (2025):

• Government ordered Tesla to fix deceptive business practice violations
• $58,000 daily fines if Tesla fails to comply
• Regulators determined “autonomous driving” features don’t meet implied standards

Germany:

• Advertising for “Autopilot” restricted as misleading
• Sales materials must clearly state driver supervision requirements

These international actions demonstrate a global regulatory consensus that Tesla’s marketing exceeds its technology’s actual capabilities.

Standard of Care for AI Driver-Assistance Systems

#

The Tesla litigation and regulatory actions establish an emerging standard of care for automakers deploying AI driver-assistance technology.

Design Standards

#

1. Capability-Appropriate Operational Domains

Manufacturers must restrict system operation to conditions where the AI can perform safely:

2. Robust Driver Monitoring

The driver monitoring system must match the system’s permissiveness:

• Eye tracking: Detect when driver looks away
• Hands-on-wheel detection: Verify physical engagement
• Escalating warnings: Progressive alerts for inattention
• Automatic disengagement: If driver fails to respond

NHTSA’s Finding: Tesla’s weak driver engagement system combined with permissive capabilities created a mismatch that is below industry standard.

3. Fail-Safe Protocols

Systems must have defined responses when they exceed operational limits:

• Graduated handoff procedures
• Safe stop capabilities
• Clear communication of system status

Disclosure Standards

#

1. Accurate Naming

System names must reflect actual capabilities:

• Avoid terms implying higher autonomy than provided
• Use SAE level designations consistently
• Ensure marketing aligns with owner’s manual disclosures

2. Clear Limitations

Every user interaction should reinforce:

• System limitations in specific scenarios
• Required driver attention levels
• Known failure modes

3. Ongoing Communication

Manufacturers must update owners about:

• Software changes affecting behavior
• Newly discovered limitations
• Recommended use restrictions

Testing and Validation Standards

#

1. Pre-Deployment Testing

• Comprehensive simulation covering edge cases
• Real-world testing in intended operational domains
• Independent safety assessment before feature releases

2. Post-Deployment Monitoring

• Real-world performance tracking
• Incident investigation protocols
• Rapid response to identified safety issues

3. OTA Update Governance

• Regression testing before deployment
• Staged rollouts with monitoring
• Rollback capability for problematic updates

Data Logging Standards

#

1. Comprehensive Event Recording

Systems should log:

• System state at time of incidents
• Driver engagement status
• Sensor inputs and AI decision points
• Warning issuance and driver response

2. Preservation and Access

• Data retention sufficient for litigation/investigation
• Owner access to their own vehicle data
• Regulatory access for safety investigations

3. Transparent Reporting

• Timely incident reporting to regulators
• Consistent methodology across reports
• No selective disclosure

Practical Guidance for Automotive Professionals

#

For Engineers and Safety Teams

#

Design Phase:

• Map system capabilities to specific operational domains
• Design driver monitoring to match system permissiveness
• Build in fail-safe modes for edge cases
• Document design decisions and safety rationales

Testing Phase:

• Develop comprehensive edge case scenarios
• Include diverse environmental and demographic conditions
• Conduct independent safety reviews before release
• Establish clear go/no-go criteria

Post-Deployment:

• Monitor real-world performance continuously
• Investigate all incidents, not just crashes
• Update operational domain restrictions based on field data
• Communicate limitations discovered post-launch

For Legal and Compliance Teams

#

Marketing Review:

• Ensure feature names don’t imply capabilities beyond actual performance
• Review all marketing materials against owner’s manual disclosures
• Document marketing approval decisions

Regulatory Compliance:

• Track state and federal requirements across jurisdictions
• Maintain records demonstrating compliance
• Prepare for regulatory inquiries with comprehensive documentation

Litigation Preparedness:

• Preserve all design, testing, and incident data
• Maintain chain of custody for vehicle data
• Develop defensible positions based on demonstrated standard of care compliance

For Corporate Leadership

#

Risk Assessment:

• Evaluate gap between marketing claims and actual capabilities
• Quantify litigation exposure from current practices
• Assess regulatory risk in key markets

Strategic Decisions:

• Consider Level 3+ systems that acknowledge manufacturer liability
• Evaluate whether market differentiation through aggressive naming is worth legal risk
• Build safety culture that prioritizes realistic capability claims

The Floodgates Open: Post-Verdict Litigation

#

Settlements and Pending Cases

#

Following the August verdict, Tesla has settled additional cases to avoid trial:

• September 2025: Tesla settled a wrongful death lawsuit from a 2019 California crash that killed a 15-year-old boy while Autopilot was engaged
• Active cases: Approximately 12-15 pending lawsuits involve fatal or injurious crashes where Autopilot or FSD was engaged

Class Action Certification

#

In August 2025, U.S. District Judge Rita Lin certified class action status for claims that Tesla:

• Lacked the hardware to achieve promised autonomy levels
• Failed to “demonstrate a long-distance autonomous drive with any of its vehicles” despite years of marketing claims

The class potentially includes all purchasers of Tesla’s Full Self-Driving package who were promised capabilities the vehicles could not deliver.

The Death Toll

#

Over 50 deaths have been linked to crashes involving Tesla Autopilot or FSD systems. While courts traditionally sided with Tesla, arguing human error remained the primary cause, the Benavides verdict establishes that juries will hold manufacturers liable when:

• The AI system contributed to the crash
• The company knew of system limitations
• Marketing created unreasonable user expectations

The SAE Autonomy Levels: A Liability Framework

#

Understanding liability requires understanding the six SAE (Society of Automotive Engineers) levels of driving automation:

Levels 0-2: Driver Support Systems

#

• Level 0: No automation (warnings only)
• Level 1: Single function automation (cruise control OR steering)
• Level 2: Combined automation (cruise control AND steering)

Liability framework: The human driver remains legally responsible. These are “driver support” features, the driver must remain engaged and in control at all times. Tesla Autopilot and most current consumer systems operate at Level 2.

Levels 3-5: Automated Driving Systems

#

• Level 3: Conditional automation (system drives in limited conditions; human must be ready to intervene)
• Level 4: High automation (system handles all driving in defined operational domains; no human intervention required)
• Level 5: Full automation (no human needed under any conditions)

Liability framework: When the automated system is engaged within its operational design domain, primary liability shifts to the manufacturer. The human may retain secondary liability for failure to resume control when prompted (Level 3) or bear no driving responsibility at all (Levels 4-5).

The Level 3 “Grey Zone”

#

Level 3 creates particular legal complexity. As automotive analysts note: “What makes systems Level 3 is the carmaker’s willingness to assume liability for what the system does while controlling the vehicle.”

Currently, very few Level 3 systems are commercially available:

• Mercedes-Benz Drive Pilot: Available by subscription on 2025 EQS and S-Class in California and Nevada only
• BMW Personal Pilot L3: Available only in Germany

The scarcity of Level 3+ systems reflects manufacturers’ reluctance to accept the liability that comes with true autonomous operation.

NHTSA Regulatory Framework (2025)

#

The April 2025 AV Framework

#

On April 24, 2025, the Department of Transportation announced the NHTSA Autonomous Vehicle Framework under Secretary Sean Duffy. The framework establishes three key principles:

1. Safety prioritization for ongoing AV operations on public roads
2. Innovation enablement by removing unnecessary regulatory barriers
3. Commercial deployment to enhance safety and mobility

Initial Actions

#

Expanded Exemption Program: NHTSA expanded its Automated Vehicle Exemption Program to include domestically produced vehicles. Previously, only foreign AVs were eligible, disadvantaging American manufacturers.

First Exemption Issued: On August 6, 2025, NHTSA issued its first exemption for Zoox driverless vehicles under the expanded program.

Proposed FMVSS Amendments: NHTSA proposed rulemakings to amend Federal Motor Vehicle Safety Standards for vehicles with automated driving systems and no manual controls, addressing transmission, visibility, and lighting standards.

Addressing the Patchwork Problem

#

A stated goal of the federal framework is to “mitigate the risks posed by a patchwork of state laws”, a critical issue for the industry.

The State Regulatory Patchwork

#

50 States, 50 Approaches

#

In the absence of comprehensive federal regulation, over 35 states have enacted their own autonomous vehicle laws. In 2025 alone, 25 states introduced 67 bills related to autonomous vehicles.

Key State Variations

#

Permissive States (Arizona, Nevada, Texas):

• Allow fully driverless vehicles
• Require safety and reporting protocols
• Texas defines the vehicle owner as the legal “operator” rather than a human driver

Restrictive States (New York):

• Require a licensed human driver during testing
• Limit operational domains

Emerging Frameworks (California):

• The California DMV is overhauling its regulations to create “the nation’s most comprehensive rules for the operation of autonomous vehicles”
• New law effective July 2026 allows police to issue “notices of autonomous vehicle noncompliance” to companies when their vehicles break traffic rules

The Compliance Burden

#

This creates what experts call a “state-by-state slog”. Companies testing in multiple states face entirely different requirements in each jurisdiction, increasing costs and diverting resources from technology development.

Industry Advocacy

#

The Autonomous Vehicle Industry Association (AVIA) released its 2025 policy framework calling for federal preemption to counter regulatory uncertainty. Currently, 25 states have adopted AV statutes that vary significantly in requirements for testing, deployment, and reporting.

California Collision Data: The National Testing Ground

#

Reporting Requirements

#

California requires autonomous vehicle operators to report any collision, no matter how minor, to the DMV. This creates the most comprehensive dataset on AV incidents available.

Current Statistics

#

As of November 2025, the California DMV has received 894 Autonomous Vehicle Collision Reports.

National Context (NHTSA Data):

• Between June 2024 and March 2025: 570 reported crashes involving cars with automated driving systems
• Monthly crashes increased from 42 in June 2024 to 81 in December 2024
• California leads with 1,120 total reported incidents

Crash Rate Comparisons

#

A 2022 analysis found:

• AVs: 96.7 crashes per 1,000 vehicles; 26.3 per million miles
• Human-driven vehicles: 7.0 crashes per 1,000 vehicles; 0.7 per million miles

Important Context: AV companies must report any collision, including parking lot bumps and low-speed contacts. Human-driven crash data relies on police reports and insurance claims, minor incidents go unreported. The comparative crash rates likely overstate AV risk.

Additionally, injuries in AV crashes tend to be less severe than in human-driven crashes.

Waymo and Cruise: Robotaxi Liability

#

Waymo’s Position

#

Waymo operates fully driverless Level 4 robotaxis in San Francisco, Los Angeles, Phoenix, and Austin. Unlike Tesla, Waymo assumes manufacturer liability, there is no human driver to blame.

Safety Claims: A peer-reviewed Swiss Re study found Waymo vehicles reduced property damage claims by 76% and eliminated bodily injury claims entirely compared to human drivers over 3.8 million miles.

Federal Investigation: Despite these claims, NHTSA opened an investigation in 2024 into 22 separate Waymo incidents, with 17 involving apparent disobedience of traffic control devices.

Cruise Suspension

#

General Motors’ Cruise subsidiary was banned from San Francisco operations in October 2023 after an incident where a Cruise vehicle ran over a pedestrian already struck by another car, stopping with its wheel pinning the woman’s leg to the ground.

The California DMV revoked Cruise’s permit, citing safety concerns and alleged misrepresentation of the incident to regulators. As of late 2025, Cruise operations remain suspended.

Public Hostility

#

Autonomous vehicles have faced growing public opposition:

• February 2024: A Waymo vehicle was surrounded by a crowd in San Francisco, had its windows smashed, and was set ablaze
• June 2025: Five Waymo vehicles were torched during protests in Los Angeles, leading to service suspension in downtown LA

Enforcement Gaps

#

Current regulations create accountability gaps. When a robotaxi violates traffic laws, there is no human driver to cite. As San Bruno police noted after a Waymo made an illegal U-turn: “Our citation books don’t have a box for ‘robot.’”

California’s 2026 law will allow “notices of autonomous vehicle noncompliance” to companies, though penalties remain under development.

Emerging Liability Theories

#

Algorithmic Negligence

#

Traditional negligence requires proving a duty of care, breach of that duty, and causation. For autonomous vehicles, this framework is evolving to address AI decision-making:

Data-Driven Negligence: Liability may shift from mechanical error to data-driven negligence when AI algorithms cause misjudgment or delayed responses.

The Black Box Problem: AI systems generate decisions that engineers cannot fully explain. Courts may reason that “using a self-learning algorithm in a sensitive context is a choice with foreseeable hazards”, even if specific emergent behaviors weren’t anticipated.

Over-the-Air (OTA) Update Liability

#

Modern vehicles receive software updates remotely, creating new liability questions:

• Flawed updates: Did the manufacturer push a software update that introduced a defect?
• Failure to update: Did the manufacturer fail to deploy a safety update?
• Unauthorized modifications: Did the user alter software in ways that voided safety protections?

The UK’s Automated and Electric Vehicles Act 2018 explicitly addresses liability for accidents “resulting from unauthorized software alterations or failure to update software.”

Duty to Provide Continuous Safety Updates

#

An emerging question: Do manufacturers have an ongoing duty to provide safety updates throughout a vehicle’s lifetime? If a known vulnerability exists and no patch is issued, has the manufacturer breached its duty of care?

This parallels cybersecurity liability debates, as AI systems require ongoing maintenance, the scope of manufacturer responsibility extends well beyond the point of sale.

Marketing and Disclosure Liability

#

The Benavides verdict highlights liability for misleading marketing:

• Capability overstatement: Describing Level 2 systems with terms suggesting higher autonomy (“Autopilot,” “Full Self-Driving”)
• Safety misrepresentation: Claims that AI systems are “safer than human drivers” without adequate disclosure of limitations
• Failure to warn: Inadequate communication of operational design domain restrictions

Product Liability Framework

#

Design Defect Claims

#

Plaintiffs in AV cases typically allege design defects:

1. Unreasonable risk: The AI system creates risks that outweigh its benefits
2. Feasible alternatives: Safer design alternatives existed and weren’t implemented
3. Failure to meet consumer expectations: The system doesn’t perform as a reasonable consumer would expect given its marketing

Manufacturing Defect Claims

#

These arise when individual units deviate from design specifications:

• Sensor miscalibration
• Software deployment errors
• Hardware component failures

Failure to Warn Claims

#

Manufacturers must provide adequate warnings about:

• Operational limitations
• Environmental conditions affecting performance
• Required driver attention levels
• Known failure modes

Insurance Implications

#

Coverage Uncertainty

#

The AI insurance crisis particularly affects autonomous vehicles:

• Traditional auto policies assume human drivers
• Product liability coverage may not extend to AI decision-making
• Lines between driver negligence and product defect blur

Emerging Coverage Models

#

Some insurers are developing AV-specific products:

• Per-mile coverage that adjusts based on autonomy level engaged
• Manufacturer-backed insurance where OEMs assume liability
• Fleet coverage for robotaxi operators

Data as Evidence

#

AV litigation requires technical proof from:

• System logs documenting AI decisions
• Update histories showing software versions
• Sensor data recording environmental conditions
• Disengagement reports when human drivers took over

Preserving this data is critical for both plaintiffs and defendants.

The Emerging Standard of Care

#

For Manufacturers

#

Based on litigation trends and regulatory developments, the emerging standard requires:

System Design:

• Adequate operational design domain restrictions
• Robust driver monitoring for Level 2 systems
• Failsafe protocols when conditions exceed system capabilities

Disclosure:

• Clear communication of system limitations
• Accurate marketing that doesn’t overstate capabilities
• Ongoing updates about known issues

Monitoring:

• Real-world performance tracking
• Incident investigation and reporting
• Continuous improvement based on field data

Updates:

• Timely deployment of safety patches
• Clear communication about update content
• Backward compatibility for older hardware

For Vehicle Owners

#

Owners of vehicles with driver-assistance systems should:

• Understand the actual autonomy level of their system
• Remain attentive regardless of marketing claims
• Install software updates promptly
• Document any incidents or unexpected behavior
• Preserve vehicle data if involved in a crash

For Attorneys

#

Plaintiff’s Counsel:

• Preserve all vehicle data immediately after incidents
• Engage technical experts who can analyze AI decision logs
• Investigate marketing claims versus actual capabilities
• Consider class action potential for systemic defects

Defense Counsel:

• Document driver behavior and attention
• Establish operational design domain boundaries
• Demonstrate regulatory compliance
• Preserve evidence of adequate warnings

Looking Forward

#

Key Questions for Courts

#

• Does using a Level 2 system shift any liability from driver to manufacturer?
• What disclosure duties exist for AI limitations not discovered until after sale?
• Can manufacturers disclaim liability through terms of service?
• How do punitive damages apply to corporate AI decisions?

Regulatory Trajectory

#

The federal government appears committed to enabling AV deployment while establishing safety baselines. State regulations will likely converge as federal standards develop, though the timeline remains uncertain.

Technology Evolution

#

As vehicles approach true Level 4+ autonomy, liability will increasingly rest with manufacturers rather than human operators. Companies accepting this liability shift, like Waymo, may gain competitive advantage through clearer accountability.

Resources

#

• California DMV Autonomous Vehicle Collision Reports
• NHTSA AV Framework Announcement
• SAE Levels of Driving Automation
• Tesla Benavides Verdict Coverage (CNN)
• Waymo Safety Impact Research
• AVIA Federal Policy Framework

Related Sites

#

If you’ve been injured in an autonomous vehicle accident, these resources may help:

• Autonomous Vehicle Injuries & Liability, Comprehensive guide to AV accident claims and legal options
• Find an AV Injury Law Firm, Directory of law firms handling autonomous vehicle accident cases