The first place to look after a defective-part incident is usually not an artificial-intelligence statute. It is the purchase order and related contracts governing the automotive supply relationship.
Michigan commercial law has long distinguished between tort claims involving personal injury and other traditional tort harms and commercial disputes involving disappointed expectations concerning purchased goods. In Neibarger v Universal Cooperatives, Inc., the Michigan Supreme Court adopted the economic-loss doctrine and held that when a commercial purchaser seeks recovery for economic loss caused by defective goods, the purchaser generally must look to the remedies provided by the Uniform Commercial Code rather than recasting the dispute as a tort action. ¹
That principle can be highly important in an automotive quality dispute. A Tier Two supplier may promise a Tier One customer that every delivered component will conform to drawings, specifications, quality requirements, engineering standards, approved samples, regulatory requirements, and the purchase order. A Tier One supplier may make similar promises to an OEM. The fact that the supplier decided to use an AI inspection system to help satisfy those obligations generally does not change what the supplier promised to deliver.
If the component fails to meet the contractual specifications, the customer may therefore have a breach-of-contract or breach-of-warranty claim against the supplier regardless of whether an employee, conventional camera, coordinate-measuring machine, statistical sampling process, or sophisticated AI model failed to detect the problem.
The supplier’s use of AI is more likely to become relevant in determining whether the supplier has a separate claim against the company that provided the inspection technology.
This distinction should be understood before an AI quality system is purchased. The downstream contract may impose almost absolute economic responsibility for nonconforming parts while the upstream AI contract offers only limited promises concerning the accuracy of the inspection technology. If that happens, the supplier can be left in the middle, owing millions of dollars to its automotive customer while possessing only a substantially smaller contractual claim against its technology provider.
Michigan’s Uniform Commercial Code recognizes express and implied warranties in transactions involving goods. An express warranty can arise from an affirmation of fact, promise, description, sample, or model that becomes part of the basis of the bargain. Michigan law also recognizes an implied warranty of merchantability in appropriate sales transactions and an implied warranty of fitness for a particular purpose when the seller knows the buyer’s particular purpose and knows that the buyer is relying upon the seller’s skill or judgment. ²
Those rules can matter on both sides of an AI quality-control dispute.
When a Michigan automotive supplier sells a component, drawings and engineering specifications may become central evidence of what was promised. If the component was required to have a certain material composition, dimensional tolerance, weld penetration, torque specification, hardness, surface condition, or other characteristic, the dispute may begin with a comparatively simple question: did the part conform?
The AI vendor’s obligations may be much less obvious.
Suppose the AI vendor’s sales proposal states that its system “detects production defects with 99.8% accuracy.” A different vendor promises only that its software will “assist operators in identifying potential anomalies.” Another contract states that the technology is provided “as is” and that the customer remains solely responsible for all quality-control decisions.
Those three arrangements can produce dramatically different litigation.
The supplier that purchases an AI inspection system should therefore pay close attention to the difference between marketing language and enforceable contractual commitments. If detecting particular defects is commercially critical, the agreement should identify what the system is expected to detect, how performance will be measured, what constitutes a false negative, what validation procedure will be used, what operating conditions must exist, what accuracy level is promised, and what happens if the system fails to meet those requirements.
Otherwise, after a loss occurs, the parties may spend substantial resources litigating what the AI vendor actually promised.
Another complication is that AI quality-control systems often do not fit neatly into traditional commercial categories.
A system may include industrial cameras, lighting equipment, sensors, processors, networking hardware, proprietary software, cloud-based analytics, model training, installation, engineering services, ongoing calibration, technical support, and software-as-a-service subscriptions. The legal treatment of the transaction can therefore depend on its predominant character.
Neibarger recognized Michigan’s use of the predominant-purpose analysis for mixed contracts involving both goods and services. ¹ If the essence of a particular AI transaction is the sale of cameras, computers, and inspection equipment with incidental installation services, Article 2 of the UCC may have a strong claim to application. If the transaction primarily involves cloud software, consulting, model development, and ongoing analytical services, the analysis may be different.
That classification matters because the UCC brings with it important rules concerning warranties, remedies, notice, consequential damages, and limitations periods. A supplier should therefore determine during contract negotiation—not after an eight-figure quality incident—what law the parties expect to govern the relationship.
In automotive manufacturing, the cost of the defective component itself is frequently the smallest part of the dispute.
A fifty-dollar component that causes an assembly line to stop can generate losses far beyond the purchase price. The customer’s claimed damages may include inspection costs, expedited transportation, third-party sorting, rework, scrap, overtime, replacement parts, engineering expenses, administrative charges, lost production, customer penalties, warranty claims, vehicle repairs, dealer expenses, field-service actions, recalls, and other downstream costs.
Michigan’s UCC permits an appropriate buyer to recover damages resulting from nonconforming accepted goods and, in proper cases, incidental and consequential damages. Consequential damages may include losses associated with particular requirements or needs that the seller had reason to know at the time of contracting and that could not reasonably have been prevented. ³
That creates a critical mismatch risk in AI contracting.
Assume an automotive supplier’s customer seeks $4 million arising from defective components that escaped inspection. The supplier’s purchase order with the customer permits recovery of line shutdown charges, sorting costs, warranty expenses, recall costs, and other consequential losses. The AI vendor agreement, however, disclaims consequential damages and caps the AI company’s total liability at the amount of subscription fees paid during the preceding twelve months.
The automotive supplier may successfully establish that the AI system failed and still recover only a small fraction of its downstream exposure.
For that reason, manufacturers evaluating AI quality technology should negotiate risk allocation with the entire supply chain in mind. An AI agreement should not be reviewed as though it were merely another software subscription. If the system is being placed in a quality gate where failure can expose the manufacturer to substantial automotive claims, the liability provisions deserve attention comparable to the technical specifications.
Michigan’s UCC generally permits commercial parties to establish limited remedies and, subject to applicable legal restrictions, to limit or exclude consequential damages. A contractual remedy may also become problematic if it fails of its essential purpose.⁴
Those concepts can become decisive when an AI system misses a defective automotive component.
Technology agreements frequently contain limitations that are routine in the software industry but potentially dangerous in a manufacturing environment. An agreement may disclaim implied warranties, promise only commercially reasonable efforts, exclude lost profits and production losses, exclude recall or replacement costs, limit claims to direct damages, impose an aggregate liability cap, or make service credits the purchaser’s exclusive remedy.
For an ordinary administrative software platform, those provisions may reflect a commercially understandable allocation of risk. For technology controlling a critical inspection point on an automotive production line, the same provisions can create a substantial exposure gap.
A manufacturer should therefore ask a practical question before signing: if the system fails in precisely the way we are purchasing it to prevent, what can we actually recover?
The answer should be determined from the contract rather than assumed from the vendor’s sales presentation.
The automotive industry presents an additional complication because the supplier’s obligations to its customer may be unusually broad.
Major automotive purchasing arrangements commonly contain detailed provisions addressing quality assurance, nonconforming goods, warranties, recalls, field actions, indemnification, and related financial responsibility. Ford Motor Company’s publicly available Production Purchasing Global Terms and Conditions, for example, contain separate sections addressing quality assurance, delivery of nonconforming goods, warranty, recalls and field service actions, and supplier indemnification.⁵
The significance is not that every supplier has identical obligations. They do not. The significance is that automotive purchasing relationships frequently allocate risks through a network of incorporated terms, supplier manuals, quality requirements, purchase orders, engineering specifications, warranty programs, and indemnification provisions.
A supplier evaluating an AI system should consequently compare the AI vendor’s contract directly against the supplier’s obligations to its automotive customer.
If the supplier promises its customer compliance with specifications, broad warranty protection, recall responsibility, defense obligations, and reimbursement for quality costs while the AI vendor promises little more than software availability, the supplier bears most of the risk even though the AI system performs the inspection.
Automotive suppliers sometimes focus initially on whether an AI vendor was “negligent.” Contractual indemnification may ultimately be more important.
Michigan automotive litigation provides examples of disputes in which contractual indemnity provisions determined responsibility for claims connected with supplier performance. In ALPS Automotive, Inc v General Motors Corp., an unpublished Michigan Court of Appeals decision involving allegedly defective automotive clock springs, the parties litigated contractual obligations concerning indemnification and notice. Although unpublished decisions must be treated accordingly and are not binding precedent in the same manner as published Michigan appellate opinions, the case illustrates how heavily automotive disputes can turn on the precise wording of indemnification agreements.⁶
An AI quality-control contract should therefore address whether the vendor will indemnify the manufacturer when a failure of the system causes or contributes to defective goods escaping inspection. It should also address whether indemnification covers only third-party personal-injury claims or extends to customer claims, recalls, property damage, warranty charges, and other commercial losses.
The word “indemnify” alone is not enough. The scope matters.
Even when everyone agrees that a defective part escaped an AI inspection system, that fact does not automatically establish that the AI vendor is responsible for the entire loss.
The AI system usually did not manufacture the defect. A machining process, casting problem, improper material, welding condition, assembly error, contamination event, tooling problem, operator error, or other production issue may have created the nonconformity. The AI system’s alleged failure may instead have been the failure to detect an already-existing defect.
That distinction can create complicated causation questions.
The vendor may argue that the inspection system was never designed to detect that category of defect. It may claim the manufacturer’s employees changed the acceptance threshold, altered lighting conditions, repositioned cameras, introduced a new part variation without retraining the model, ignored maintenance requirements, bypassed a warning, failed to install an update, or operated outside specified conditions.
The manufacturer may respond that those conditions were foreseeable, that the system was marketed for precisely that production environment, or that the vendor’s implementation engineers approved the configuration.
A further issue is whether the part would actually have been contained if the AI system had produced a different result. If the manufacturer’s process permitted employees to override rejected parts without secondary review, the vendor may argue that human conduct breaks or reduces the causal connection between the software output and the downstream shipment.
Future AI quality disputes will therefore often require more than traditional contract discovery. The evidence may include model versions, confidence scores, image files, classification thresholds, retraining records, software updates, configuration histories, operator overrides, validation studies, maintenance records, defect libraries, false-positive and false-negative data, and communications between plant personnel and vendor engineers.
The moment a significant defective-part issue arises, preservation decisions can become critical.
Traditional automotive quality investigations already collect parts, photographs, measurement data, inspection reports, PPAP materials, control plans, corrective-action records, and manufacturing information. AI introduces another evidentiary layer. The manufacturer may need to determine exactly which model version inspected the affected lot, which configuration was active, what images were captured, what result the model produced, what confidence score was assigned, whether an operator overrode the result, whether system performance had drifted, and whether software or training data changed after the affected parts were produced.
This information can disappear surprisingly quickly if the vendor automatically replaces models, overwrites logs, deletes stored images, or continuously updates cloud software.
The contract should therefore address data retention before a dispute exists. Once a material quality claim is reasonably anticipated, manufacturers and their counsel should consider whether ordinary document-preservation measures need to encompass the AI system’s technical records as well.
Without those records, both sides may be forced to reconstruct what the system did from incomplete evidence.
When defective goods create only commercial losses, Michigan’s economic-loss doctrine can substantially affect litigation strategy.
In Neibarger, the Michigan Supreme Court emphasized that commercial parties purchasing goods should generally be held to the allocation of risks established through contract and the UCC when their complaint concerns disappointed commercial expectations. ¹ The doctrine prevents a commercial purchaser from automatically avoiding contractual warranty limitations or UCC remedies simply by labeling the seller’s conduct negligence.
Applied to automotive AI, the doctrine can operate at multiple levels.
The OEM or Tier One customer may have primarily contractual claims against the component supplier. The supplier’s claim against an AI vendor may likewise be principally contractual if the transaction is governed by the UCC and the claimed injury is failure of the purchased system to perform as expected. Whether the UCC applies to the AI agreement itself, however, may depend on the nature and predominant purpose of the transaction.
This is another reason why the contract’s governing-law provisions, warranties, exclusions, limitations, and remedy structure should be examined before implementation.
The situation changes considerably if a defective automotive component does more than cause economic loss.
Suppose an AI system misses a critical defect in a steering component, braking component, suspension part, seat structure, airbag component, or another safety-related item. The defective part reaches a completed vehicle and contributes to an accident involving bodily injury or property damage.
Michigan’s statutory product-liability framework may then become relevant in addition to contractual obligations. Michigan law addresses production defects and permits evidence concerning generally recognized and prevailing nongovernmental production standards. For a production-defect theory, MCL 600.2946 addresses whether the product was reasonably safe when it left the manufacturer’s or seller’s control and the availability of practical and technically feasible alternative production practices.⁷
An AI inspection system could become part of that analysis.
A plaintiff might argue that the manufacturer had available technology capable of detecting the defect and failed to implement, validate, monitor, or properly maintain it. Conversely, a manufacturer could potentially point to an appropriate automated inspection system, validation procedures, industry standards, and human review as evidence of a reasonable production process.
Importantly, simply having AI does not necessarily prove reasonable care. A sophisticated inspection system that is poorly validated, operated outside its design parameters, or known to produce unacceptable false negatives could potentially create damaging evidence instead.
Technology is therefore not a substitute for a defensible quality process.
Although the legal dispute may ultimately be resolved under contract, UCC, or product-liability principles, broader AI governance can help manufacturers prevent the dispute and defend their processes if one occurs.
The National Institute of Standards and Technology’s Artificial Intelligence Risk Management Framework identifies governance, mapping, measurement, and management as core functions for organizations deploying AI systems.⁸ The framework is voluntary rather than an automotive liability statute, but its underlying approach is useful for manufacturers implementing industrial AI.
A supplier should understand what risk the AI system controls, how performance is validated, how errors are measured, who owns ongoing monitoring, when retraining is required, how changes are documented, what human review remains necessary, and how the organization responds when performance falls outside acceptable limits.
In a later lawsuit, those records may help answer an important question: did the manufacturer responsibly deploy and supervise the technology, or did it simply purchase an AI product and assume it would work?
Commercial parties also need to pay attention to notice requirements after discovering defective goods.
Michigan’s UCC requires a buyer that has accepted goods to notify the seller of breach within the applicable statutory framework, and failure to provide required notice can affect available remedies.⁹ Contracts may impose even more detailed notice obligations governing indemnification claims, quality disputes, warranty recoveries, or litigation.
This can be particularly complicated in AI-related incidents because the supplier may initially know that defective parts escaped but not yet know why.
The supplier should avoid waiting for a final root-cause determination before reviewing contractual notice obligations. Depending on the agreement, notice may be required well before the technical investigation establishes whether the AI vendor, the manufacturer’s process, an upstream material supplier, or several participants contributed to the problem.
A related trap concerns the statute of limitations.
Under Michigan’s UCC, an action for breach of a contract for sale generally must be commenced within four years after accrual, although the parties may reduce that period to not less than one year. A breach of warranty ordinarily accrues when tender of delivery occurs, even if the buyer does not yet know of the breach, subject to the statutory exception for warranties explicitly extending to future
performance. ¹⁰
That rule can matter when an AI system performs acceptably for years before historical evidence reveals that it had been systematically missing a particular defect.
A manufacturer should not assume that the limitations clock began when engineers finally discovered the AI problem. The relevant agreement and the nature of the warranty must be reviewed carefully.
The most important lesson for Michigan automotive suppliers is preventive.
Before deploying an AI system at a production quality gate, the manufacturer should negotiate the contract by imagining the worst credible failure. The agreement should clearly identify the defects the system is intended to detect, measurable acceptance criteria, responsibility for validation, operating assumptions, performance monitoring, updates, model changes, data ownership, retention of inspection records, cybersecurity responsibilities, and procedures for investigating suspected failures.
The parties should also confront financial responsibility directly. They should determine whether the vendor’s warranties actually cover detection performance, whether consequential damages are excluded, whether recall and customer-chargeback losses are recoverable, how liability caps operate, whether certain claims are carved out of those caps, what indemnification is available, what insurance must be maintained, and whether the vendor must cooperate in customer or regulatory investigations.
The manufacturer should then compare those terms with its own contractual obligations to the OEM or Tier One customer.
The goal is not necessarily to force every dollar of downstream liability onto the AI provider. A technology vendor reasonably may refuse to insure manufacturing risks that vastly exceed its compensation. The goal is to identify the gap consciously so the manufacturer can decide whether to negotiate different terms, purchase additional insurance, maintain redundant inspection, require human verification, limit the AI system to noncritical characteristics, or accept the remaining exposure as a business risk.
What manufacturers should avoid is discovering the gap only after the recall begins.
AI quality-control technology can substantially improve automotive manufacturing, but it does not eliminate legal responsibility. Instead, it redistributes responsibility among manufacturers, suppliers, technology vendors, system integrators, employees, insurers, and customers.
The supplier remains responsible for understanding what it has promised its customer. The AI vendor is responsible for whatever performance it has actually promised the supplier. Engineers and quality personnel remain responsible for validating and monitoring the manufacturing process. Lawyers and purchasing personnel should ensure that contractual risk allocation corresponds to the operational importance of the system.
When a defective component escapes an AI inspection system, there may therefore be no single answer to the question, “Who pays?”
The automotive customer may pursue the supplier for delivering a nonconforming component. The supplier may pursue the AI vendor for breach of warranty or contract. An AI vendor may invoke a liability cap or consequential-damage exclusion. An insurer may become involved. If bodily injury or property damage occurs, tort and product-liability claims may expand the dispute further. Multiple parties may ultimately share responsibility depending on contractual language and causation.
What is clear is that calling the inspection system “AI” does not itself determine liability.
For Michigan automotive manufacturers and suppliers, the better question is more precise: what did each party promise, what failed, what caused the loss, what damages are recoverable, and who contractually agreed to bear that risk?
Those questions should be answered before the first defective part passes through the camera.
Contact Tishkoff
Tishkoff PLC specializes in business law and litigation. For inquiries, contact us at www.tish.law/contact/. & check out Tishkoff PLC’s Website (www.Tish.Law/), eBooks (www.Tish.Law/e-books), Blogs (www.Tish.Law/blog) and References (www.Tish.Law/resources).
Sources:
1- Neibarger v Universal Cooperatives, Inc., 439 Mich 512; 486 NW2d 612 (1992). https://law.justia.com/cases/michigan/supreme-court/1992/88206-3.html
2- MCL 440.2313; MCL 440.2314; MCL 440.2315; MCL 440.2316. https://codes.findlaw.com/mi/chapter-440-uniform-commercial-code/mi-comp-laws-440-2313/
https://legislature.mi.gov/Laws/MCL?objectName=mcl-440-2314
https://www.legislature.mi.gov/Laws/MCL?objectName=mcl-440-2315
https://legislature.mi.gov/Laws/MCL?objectName=MCL-440-2316
3- MCL 440.2714; MCL 440.2715. https://legislature.mi.gov/Laws/MCL?objectName=mcl-440-2714 https://legislature.mi.gov/Laws/MCL?objectName=MCL-440-2715
4- MCL 440.2719. https://codes.findlaw.com/mi/chapter-440-uniform-commercial-code/mi-comp-laws-440-2719/
5- Ford Motor Company, Production Purchasing Global Terms and Conditions, provisions concerning Quality Assurance, Delivery of Nonconforming Goods, Warranty, Recalls and Other Field Service Actions, and Supplier Indemnification Obligations. https://corporate.ford.com/operations/governance-and-policies/production-purchasing-global-terms-and-conditions/
6- ALPS Automotive, Inc v General Motors Corp, unpublished per curiam opinion of the Michigan Court of Appeals, issued January 10, 2006 (Docket No. 256597). https://law.justia.com/cases/michigan/court-of-appeals-unpublished/2006/20060110-c256597-64-256597-opn.html
7- MCL 600.2945; MCL 600.2946; MCL 600.2947. https://www.legislature.mi.gov/Laws/MCL?objectName=mcl-600-2945 https://www.legislature.mi.gov/Laws/MCL?objectName=mcl-600-2946 https://www.legislature.mi.gov/Laws/MCL?objectName=mcl-600-2947
8- National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (January 2023). https://kpmg.com/us/en/articles/2026/tmt-managed-services-ai.html?utm_source=google&utm_medium=cpc&utm_content=search_text_ad&utm_campaign=701dV00000cI53OQAS&cid=701dV00000cI53OQAS&gclsrc=aw.ds&gad_source=1&gad_campaignid=24194318152&gbraid=0AAAAADq0WlrifSNa6lApDj_2CYI2pn6Gn&gclid=EAIaIQobChMI7MqC_-fflgMVHz0IBR34oTgKEAAYASAAEgL9M_D_BwE
9- MCL 440.2607. https://www.legislature.mi.gov/Laws/MCL?objectName=mcl-440-2607
10- MCL 440.2725.
https://legislature.mi.gov/Laws/MCL?objectName=MCL-440-2725