Publication date: November 17, 2026
Legal status as of 12 November 2026
The automotive industry is increasingly turning to specialised companies from the artificial intelligence sector, and AI functions are moving from cloud servers to the vehicle’s own computers. This trend is currently most visible in the US market. From a Polish perspective, embedded in the EU legal order, the key question is whether existing rules allow such solutions to be deployed commercially and whether they support cooperation between manufacturers and academic spin-outs. The answer is yes, but on-device processing is not a regulatory shortcut: it changes how obligations are distributed, not whether they exist.
Cars have long been largely software-defined products. What is new is the extent to which manufacturers (OEMs) rely on AI models developed outside their own organisations, most often by young, narrowly specialised technology companies. One example is the multi-year partnership announced in 2026 between Mercedes-Benz and Liquid AI, a company with roots at MIT, concerning embedded in-vehicle intelligence, in particular on-device speech processing and language understanding.[1] Such arrangements are becoming more common, and their common denominator is moving part of the AI computation into the vehicle itself (so-called edge AI).
The uses of AI in a car are not limited to the driver-vehicle relationship. They can be grouped as follows:
• user interaction – voice assistants, control of cabin settings, media and navigation;
• driver monitoring – detection of drowsiness and distraction and, going forward, assessment of the driver’s physical and mental condition;
• driver assistance and safety – speed control, hazard warnings, response to changing weather conditions, emergency manoeuvres;
• vehicle management – coordination of components, management of energy, fuel and fluids, predictive diagnostics;
• data-driven services – telematics, usage-based insurance, servicing and fleet management.
This map matters legally. Most obligations depend not on whether a system “is AI”, but on what it does, what it can access and what consequences its correct or faulty operation may have. Publicly described deployments of on-device language models concern mainly the first group, i.e. the in-cabin assistant, and are expressly distinguished from automated driving projects.[2] No conclusions about all vehicle functions should therefore be drawn from them. The legal assessment must be carried out function by function.
Edge AI means performing computation close to the data source, here on the vehicle’s computer. A model may be trained in a data centre, while the car performs only inference, i.e. uses the trained model to produce an output. On-device operation also does not automatically mean that the system learns from every conversation with the driver.
From a business perspective, moving computation into the vehicle brings independence from network coverage, lower latency and lower infrastructure costs. In practice, however, most deployments remain hybrid: simple commands are handled locally, while queries beyond the capabilities of a smaller model are sent to the cloud. “Offline” therefore describes the connectivity of a specific function, not a guarantee that no data will ever leave the car.
From a legal perspective, the location of computation affects data flows, transfers to third countries, exposure to incidents and the organisation of updates. It does not, however, replace an analysis of the system’s intended purpose. It is also worth distinguishing the language model from the integration layer: a well-designed assistant does not control actuators directly but selects a command from a closed catalogue, while an intermediate layer checks permissions and vehicle status and may request confirmation or refuse execution. Such an architecture makes subsequent legal classification and demonstration of compliance easier.
The AI Act[3] distinguishes between an AI model, an AI system and the product in which the system is embedded. The model is a component; the system combines it with an interface, data and tools serving a specific use; the vehicle is a still broader product subject to a separate type-approval regime. Classification at one level does not determine the others.
The provider of the system within the meaning of the AI Act will usually be the car manufacturer placing it on the market under its own brand, even if the model was supplied by a start-up. The start-up may at the same time remain the provider of the model and, if the model meets the criteria of a general-purpose AI (GPAI) model under Article 3(63) of the AI Act, it is subject to the obligations under Article 53 (technical documentation, information for downstream integrators, a copyright compliance policy and a summary of training content). The label “foundation model” in marketing materials does not determine GPAI status, and a small on-device model may well have it. A private driver using the system purely for personal purposes is not a deployer (Article 2(10) of the AI Act); a fleet operator may be.
An assistant does not become a high-risk system merely because it is installed in a vehicle. Article 6(1) of the AI Act links high risk to performing a safety function of a product covered by Union harmonisation legislation. Following the 2026 amendment,[4] the provision more clearly distinguishes functions serving solely user convenience from functions whose failure endangers health and safety.
A sectoral route applies to the automotive industry. Vehicles covered by Regulations 2018/858 and 2019/2144 fall under Section B of Annex I, to which, under Article 2(2) of the AI Act, only selected provisions apply, while requirements for AI systems are to be introduced into type-approval legislation (Articles 107-109). There is therefore no basis for attributing to every automotive safety system the full set of obligations under Chapter III of the AI Act or CE marking; a CE mark is not vehicle type-approval. At the same time, an ordinary assistant that does not fall within this exception is subject to the general provisions of the AI Act, and the exception for an automotive system does not automatically exempt the provider of a GPAI model.
In practice, a conversation about music, air-conditioning control, navigation and commands affecting functions relevant to driving should be assessed separately. Even a comfort function requires safeguards if activating it in a particular vehicle state could create a hazard (e.g. fogged windows or reduced visibility).
Where a system interacts with humans, the provider must ensure that users are informed that they are interacting with AI at the latest at the first interaction, unless this is obvious from the context (Article 50(1) of the AI Act). This does not require a notice before every response. Good practice is to separate three messages: “you are talking to AI”, “this function requires an internet connection” and “data will be used for an additional purpose”, as each has a different legal basis. Separately, Article 4 of the AI Act requires providers and deployers to take measures to ensure AI literacy of their staff, which in the automotive context covers integration, service and complaints-handling teams.
In the EU, driver monitoring is not merely a design choice. Regulation 2019/2144[5] requires new vehicles to be equipped, among other things, with driver drowsiness and attention warning (DDAW) and advanced driver distraction warning (ADDW) systems, whose technical requirements are laid down in delegated regulations.[6] By their nature, these systems process images or signals relating to a specific person and increasingly rely on machine learning.
The line runs between assessing a physical state and recognising emotions. Article 5(1)(f) of the AI Act prohibits inferring the emotions of a natural person in the workplace, except for systems intended for medical or safety reasons. A company car, a courier fleet or a carrier’s vehicle may be a workplace. Recital 18 of the AI Act clarifies that detecting physical states such as fatigue is not emotion recognition. Extending a simple drowsiness detection system to analyse the driver’s frustration or aggression therefore requires a fresh assessment rather than reliance on the previous classification.
Under the GDPR,[7] data from a driver-facing camera does not automatically become special-category biometric data (Article 9) unless it is processed for the purpose of uniquely identifying a person. Systematic monitoring of driver behaviour, especially in an employment relationship, will however very likely require a data protection impact assessment (Article 35).
The most sensitive business area is the use of driving-behaviour data by insurers. The AI Act classifies as high-risk the risk assessment and pricing of life and health insurance (Annex III, point 5(c)), but not motor insurance. This does not mean a free hand: pricing based on driver profiling is subject to the GDPR, including Article 22 on decisions based solely on automated processing, and the Data Act gives the user of a vehicle the right to share data with a chosen third party, e.g. an insurer (Article 5).[8] The US example shows the scale of the risk: in 2025 the Federal Trade Commission challenged a manufacturer’s practice of passing location and driving-behaviour data to data brokers used by insurance companies.[9] In the European model, a transparent legal basis for sharing data, separate from the assistant function itself, is therefore essential.
Moving computation into the car can reduce the transmission of voice recordings, location and command history to external servers. It is therefore a useful measure of data protection by design and by default (Article 25 GDPR). In its guidelines on connected vehicles, the European Data Protection Board expressly recommends processing data within the vehicle wherever possible.[10] The guidelines do not, however, mean that any cloud is impermissible or that every on-device solution is lawful.
“Offline” does not mean “anonymous”. A voice recording, the content of a command, a profile identifier or location remain personal data even if they never leave the car’s memory. The purpose, legal basis and retention period must therefore be established. The data-flow map should also cover diagnostic logs, service access, synchronisation after connectivity is restored and the sale of a used vehicle.
Roles of the parties. A manufacturer is not the controller of all data merely because it built the car, and a model provider is not automatically a processor. What matters is actual influence over the purposes and means of processing (Article 4(7) and Articles 26 and 28 GDPR). In a purely on-device model in which the provider has no access to data, it may not process users’ personal data at all, which simplifies the contract but must be documented. The household exemption applicable to the driver does not release businesses from obligations relating to their own processing.
Legal bases. Performing the requested function, personalisation, analytics and training future model versions are separate purposes. Article 6(1)(b) GDPR covers only processing objectively necessary for the performance of a contract; listing an operation in the terms and conditions does not make it necessary. Nor can the purchaser of a vehicle validly consent on behalf of passengers and other drivers.
Cloud and transfers. In a hybrid model, some queries are sent to servers, often outside the EEA. Articles 44-49 GDPR then apply. Handling basic commands locally reduces the scope of transfers but does not eliminate them. It is advisable to transmit only the information needed for a specific query rather than the conversation history or vehicle profile.
Model training. According to EDPB Opinion 28/2024, a model trained on personal data does not automatically become anonymous; what matters is whether data can be extracted from the parameters or obtained through the model’s outputs.[11] Deploying a ready-made model in a car does not cure defects in earlier processing, and data collected for product development requires a separate legal basis.
Retention. An on-device architecture makes it possible to keep storage to a minimum, e.g. by processing voice briefly without storing the entire conversation. Retention periods should be defined for each category of data, and when the vehicle changes hands, a simple and effective function to delete the previous user’s data is essential.
Access to information stored in the device. Independently of the GDPR, storing information in terminal equipment and gaining access to it are governed by provisions implementing Article 5(3) of the ePrivacy Directive, in Poland Article 399 of the Electronic Communications Law.[12] According to EDPB Guidelines 2/2023, an entirely local operation without access by another entity does not always fall within that regime; however, prior installation of software, updates and subsequent transmission of data are relevant.[13] A generic “I agree to AI” screen will not replace a description of the actual operations or an analysis of the strict-necessity exemption. Sending promotional content through the assistant must be assessed separately (Article 398 of the Electronic Communications Law).
Data Act. A modern car is usually a connected product, and some of its digital functions are related services. The Data Act gives the user the right to access data generated by the use of the product and to share it with third parties, such as repair shops or insurers (Articles 4-5). The obligation to design products so that data is accessible to the user by default (Article 3(1)) applies to products placed on the market after 12 September 2026. Importantly, an on-device architecture does not exclude these obligations: data stored only in the vehicle may still need to be made available to the user. The Data Act does not, however, require every conversation to be recorded or model weights to be disclosed. The Commission’s guidelines on vehicle data help distinguish raw and pre-processed data, which are subject to the access right, from inferred information.[14] In rental, leasing and fleet arrangements, it should be borne in mind that the owner of the vehicle is not the data subject for all the data (Article 4(12) of the Data Act).
Portability. The right to data portability under Article 20 GDPR and the rights under the Data Act differ in scope and addressees. In practice, it is advisable to handle both regimes through a single, consistent interface that distinguishes the driver’s personal data from product data.
Breaches. The controller notifies a personal data breach to the President of the Polish Data Protection Office (UODO) within 72 hours of becoming aware of it, unless the breach is unlikely to result in a risk to individuals (Articles 33-34 GDPR). If the provider of the assistant service is also a provider of electronic communications services, the special 24-hour regime under Article 402 of the Electronic Communications Law may apply. Equipping a car with connectivity does not in itself confer such status, however. On-device processing limits the scale of potential incidents but does not eliminate risk, e.g. in the event of physical access to the vehicle.
Type-approval. Under Regulation 2018/858,[15] the manufacturer is responsible to the approval authority (in Poland, the Transport Technical Supervision, TDT) for type-approval and conformity of production. A contract with a model provider does not transfer this public-law responsibility. AI documentation prepared by a start-up is input for integration but does not replace the information folder, technical service reports or evidence of conformity of production.
Cybersecurity. UNECE Regulation No 155[16] requires a certified cybersecurity management system (CSMS) and, for a specific vehicle type, demonstration of threat analysis, supply-chain risk management and test results. For OEMs, this means bringing the AI start-up within their own security processes, including contractual obligations concerning vulnerabilities. The Cyber Resilience Act (CRA)[17] excludes products covered by Regulation 2019/2144, but not all software of automotive suppliers: a separately offered application or module may fall under the CRA, whose vulnerability reporting obligations apply from 11 September 2026 and whose main requirements apply from 11 December 2027.
OTA updates. Edge AI requires regular model updates in vehicles already in use. UNECE Regulation No 156[18] provides for a certified software update management system (SUMS), identification of software versions, assessment of an update’s impact on the approved type, protection of integrity and safe installation conditions. In the EU, these requirements apply to all new vehicles from 7 July 2026.[19] Not every model update requires a new type-approval: whether an amendment or extension of the certificate is required depends on the impact of the change on approved characteristics (Articles 33-34 of Regulation 2018/858). The rationale for this assessment must be retained before the update is distributed. ISO 26262, ISO 21448 and ISO/SAE 21434 are voluntary, but in practice serve as contractual and evidentiary benchmarks.
The offline promise. If a manufacturer advertises certain functions as working without a connection, that representation affects the assessment of the goods’ conformity with the contract (Article 43b of the Polish Consumer Rights Act).[20] The trader must also supply updates necessary to keep the goods in conformity, including security updates, for a period determined by statutory criteria, not automatically for two years (Article 43k). An update that removes an advertised on-device function or starts transmitting additional data may result in lack of conformity and requires a fresh consumer and privacy review. Sales materials should indicate which commands work locally and which require a connection, an account or a subscription.
The seller’s liability for the car’s conformity with the contract must be distinguished from liability for damage caused by a dangerous product (Article 449¹ et seq. of the Polish Civil Code)[21] and contractual liability along the supply chain. Contractual allocation of risk between the OEM and the start-up does not limit the injured party’s rights.
A significant change will come with the new Product Liability Directive,[22] which expressly covers software, including AI systems, updates and related digital services remaining within the manufacturer’s control. The transposition deadline is 9 December 2026, and the new rules will apply to products placed on the market after that date. The Polish implementing act is at the draft stage.[23] For contracts concluded today, this means ensuring that the OEM has access to the version history, test results, vulnerability information and evidence of defect remediation, since the manufacturer will be the first target of claims, including those relating to a component supplied by a start-up. The location of computation does not in itself determine defectiveness.
An autonomous, “intelligent” vehicle requires many narrow competences: speech recognition in a noisy cabin, assessment of the driver’s condition, perception in difficult weather, emergency manoeuvre planning or energy optimisation. A single technology company is rarely able to develop all of them at the highest level. It is precisely in such niches that teams emerging from universities, after years of fundamental research in a specific field, have an advantage. It can therefore be argued that cooperation between manufacturers and academic spin-outs will become increasingly common, and that the vehicle of the future will integrate solutions from many specialised suppliers.
The US model. In the United States, commercialisation of university research relies on specialised technology licensing offices, which grant licences to spin-outs (often exclusive within a defined field of use) and may accept equity in lieu of part of the licence fees.[24] The roles of the research laboratory, the technology licensing office and the company itself are kept separate, and conflicts of interest of founders who are also university employees are subject to disclosure and management. This model ensures a clear chain of title, which is key for an industrial customer.
The Polish framework. In Poland, a spin-out is not a separate legal form; it most often takes the form of a limited liability company or a simple joint-stock company, which may be formed for any lawful purpose and without any requirement that a shareholder be a researcher (Article 151 § 1 and Article 300¹ § 1 of the Commercial Companies Code).[25] The Law on Higher Education and Science[26] distinguishes direct commercialisation, carried out by a technology transfer centre (sale of rights or licensing), from indirect commercialisation, carried out by a university special-purpose company acquiring shares in companies implementing research results (Articles 148-150a). Articles 152-157 of that Act govern research results created within the duties of a university employee: the obligation to disclose them, the university’s decision on commercialisation, the possibility of transferring rights to the creator and the sharing of proceeds. The creator’s statutory share of proceeds is not an equity stake in the spin-out.
The chain of title must be traced separately for each component of the technology. Polish copyright law[27] differentiates between employee works, scientific works (Article 14) and computer programs (Article 74(3), with a presumption that rights vest in the employer). Model weights, training datasets and know-how may be subject to regimes different from those applying to code. The lawfulness of training data requires separate analysis: the text and data mining exception for scientific research (Article 26² of the Copyright Act) is broader than the general exception (Article 26³) available to commercial entities. A company’s academic origin does not extend a research organisation’s privileges to its business activities.
In practice, the main obstacle is not a lack of legal basis but the time and uncertainty of university commercialisation procedures and an unclear chain of title. For a manufacturer or investor, due diligence on a spin-out should cover the founders’ agreements with the university, commercialisation decisions, licences and any encumbrances arising from public research funding.
Cooperation between an OEM and a start-up is rarely a single contract. It usually comprises a technology licence, a joint development agreement, a software supply and maintenance agreement and sometimes an equity investment. These must be kept distinct: a licence does not transfer shares, an investment does not confer rights to the code, and a partnership is not an acquisition. From the perspective of an edge AI deployment, the contract should at least address:
• the right to install the model on devices in a defined number of vehicles and versions, to adapt it to a specific processor and to have it used by vehicle purchasers;
• rights to further training, modifications and improvements, including the possibility of using them outside the project;
• licences for third-party components and open models, whose terms must be read rather than assumed from the label “open”;
• the parties’ roles under the GDPR and the AI Act and the scope of the start-up’s access to vehicle data;
• compliance documentation to be provided to the OEM for type-approval, CSMS and SUMS purposes, together with vulnerability reporting obligations;
• rules and timelines for updates, regression testing and preservation of advertised offline functions;
• continuity of support throughout the vehicle’s life cycle, including in the event of the start-up’s insolvency or acquisition (e.g. escrow of code and model weights);
• allocation of liability and recourse, taking into account the new product liability regime.
The US market is deploying on-device AI in vehicles quickly, partly because of its regulatory design. Federal vehicle safety standards in the US rely on manufacturer self-certification rather than prior type-approval. The absence of a general federal data protection statute, in turn, results in a patchwork of state laws and a growing role for ex post enforcement, as illustrated by the FTC intervention on driving-behaviour data mentioned above. For other jurisdictions, including the EU, three lessons follow: on-device processing is an effective tool for reducing privacy risk; monetisation of driver data is the area of highest regulatory and reputational risk; and narrowly specialised spin-outs are becoming natural partners for manufacturers.
Under Polish and EU law, commercial deployment of automotive edge AI is permissible, and this direction is in fact favoured by the principle of data protection by design. On-device solutions do not, however, release businesses from obligations under the AI Act, the Data Act, type-approval law, the CRA, consumer law and the product liability regime. They change how those obligations are distributed: they reduce exposure to cloud transfers and incidents, but increase the importance of updates, documentation and maintaining functions throughout the vehicle’s life cycle.
We assess the readiness of the Polish legal environment as high in terms of fundamentals (company law, university commercialisation, data protection), but incomplete in several areas. These include the pending transposition of the Product Liability Directive, the practice of the newly established AI supervisory bodies under the Act on Artificial Intelligence Systems,[28] further sectoral acts introducing AI requirements into type-approval and the speed of university commercialisation procedures. For businesses, this means that competitive advantage today depends not only on the quality of the model, but also on a well-designed architecture of contracts, data and compliance.
| Date | Event |
| 7 July 2026 | Software update requirements (UN R156) for all new vehicles in the EU |
| 2 August 2026 | AI Act – main application of transparency obligations (Article 50) |
| 11 September 2026 | CRA – obligation to report actively exploited vulnerabilities and severe incidents |
| 12 September 2026 | Data Act – access-by-design obligation for newly placed products |
| 9 December 2026 | Transposition deadline for the Product Liability Directive |
| 2 August 2027 | AI Act – compliance deadline for GPAI models placed on the market before 2 August 2025 |
| 11 December 2027 | CRA – main requirements apply |
| 2 August 2028 | AI Act – requirements for high-risk systems under Annex I (subject to the sectoral route for vehicles) |
[1] Mercedes-Benz Group, press release of 23 April 2026 on the collaboration with Liquid AI, https://group.mercedes-benz.com/technology/innovation/collaboration/liquid-ai.html; see also MIT CSAIL Alliances, Liquid AI, Startup Connect, https://cap.csail.mit.edu/node/4185.
[2] I. Bousquette, Mercedes Vehicles Get an AI Upgrade, The Wall Street Journal, 10 November 2026, p. B4.
[3] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), OJ L, 2024/1689, 12.7.2024, as amended (the “AI Act”).
[4] Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplification of the implementation of harmonised rules on artificial intelligence, OJ L, 2026/1744, 24.7.2026.
[5] Regulation (EU) 2019/2144 of the European Parliament and of the Council of 27 November 2019 on type-approval requirements for motor vehicles […] as regards their general safety, OJ L 325, 16.12.2019, p. 1, Article 6(1).
[6] Commission Delegated Regulation (EU) 2021/1341 (DDAW) and Commission Delegated Regulation (EU) 2023/2590 (ADDW).
[7] Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 (General Data Protection Regulation), OJ L 119, 4.5.2016, p. 1, as amended (the “GDPR”).
[8] Regulation (EU) 2023/2854 of the European Parliament and of the Council of 13 December 2023 on harmonised rules on fair access to and use of data (Data Act), OJ L, 2023/2854, 22.12.2023 (the “Data Act”).
[9] Federal Trade Commission, FTC Takes Action Against General Motors for Sharing Drivers’ Precise Location and Driving Behavior Data Without Consent, press release of 16 January 2025.
[10] EDPB, Guidelines 01/2020 on processing personal data in the context of connected vehicles and mobility related applications, version 2.0, 9 March 2021, paras 73-76.
[11] EDPB, Opinion 28/2024 on certain data protection aspects related to the processing of personal data in the context of AI models, 18 December 2024.
[12] Directive 2002/58/EC of the European Parliament and of the Council of 12 July 2002 (Directive on privacy and electronic communications), OJ L 201, 31.7.2002, p. 37, as amended; Act of 12 July 2024 – Electronic Communications Law (Prawo komunikacji elektronicznej), Journal of Laws (Dz. U.) 2024, item 1221, as amended.
[13] EDPB, Guidelines 2/2023 on the Technical Scope of Art. 5(3) of the ePrivacy Directive, version 2.0, 7 October 2024.
[14] Commission Communication – Guidelines on vehicle data, accompanying Regulation (EU) 2023/2854 (Data Act), OJ C, C/2025/5026, 15.9.2025.
[15] Regulation (EU) 2018/858 of the European Parliament and of the Council of 30 May 2018 on the approval and market surveillance of motor vehicles and their trailers […], OJ L 151, 14.6.2018, p. 1; Commission Implementing Regulation (EU) 2020/683, OJ L 163, 26.5.2020, p. 1.
[16] UN Regulation No 155 – Uniform provisions concerning the approval of vehicles with regards to cyber security and cyber security management system [2025/5], OJ L, 2025/5, 10.1.2025.
[17] Regulation (EU) 2024/2847 of the European Parliament and of the Council of 23 October 2024 (Cyber Resilience Act), OJ L, 2024/2847, 20.11.2024, Articles 2(2)(c), 14, 69 and 71.
[18] UN Regulation No 156 – Uniform provisions concerning the approval of vehicles with regards to software update and software update management system [2021/388], OJ L 82, 9.3.2021, p. 60; the 01 series of amendments applies from 4 June 2026.
[19] Commission Delegated Regulation (EU) 2022/2236 of 20 June 2022, OJ L 296, 16.11.2022, p. 1, Article 2(7).
[20] Act of 30 May 2014 on Consumer Rights (ustawa o prawach konsumenta), Journal of Laws (Dz. U.) 2026, item 1244.
[21] Act of 23 April 1964 – Civil Code (Kodeks cywilny), Journal of Laws (Dz. U.) 2026, item 795.
[22] Directive (EU) 2024/2853 of the European Parliament and of the Council of 23 October 2024 on liability for defective products, OJ L, 2024/2853, 18.11.2024; corrigendum: OJ L, 2026/90364, 7.5.2026, Articles 4, 7, 11 and 21-22.
[23] Draft Act amending the Civil Code and the Code of Civil Procedure (UC163), List of legislative works of the Council of Ministers, entry of 25 August 2026.
[24] MIT Technology Licensing Office, Understanding the Licensing Process, https://tlo.mit.edu/understand-ip/understanding-licensing-process; MIT Policies and Procedures, 13.1 Intellectual Property and 4.4 Conflict of Interest.
[25] Act of 15 September 2000 – Commercial Companies Code (Kodeks spółek handlowych), Journal of Laws (Dz. U.) 2024, item 18, as amended.
[26] Act of 20 July 2018 – Law on Higher Education and Science (Prawo o szkolnictwie wyższym i nauce), Journal of Laws (Dz. U.) 2026, item 1362.
[27] Act of 4 February 1994 on Copyright and Related Rights (ustawa o prawie autorskim i prawach pokrewnych), Journal of Laws (Dz. U.) 2025, item 24, as amended (the “Copyright Act”); Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019, OJ L 130, 17.5.2019, p. 92, Articles 3-4.
[28] Act of 3 July 2026 on Artificial Intelligence Systems (ustawa o systemach sztucznej inteligencji), Journal of Laws (Dz. U.) 2026, item 1003.
KIELTYKA GLADKOWSKI is a Polish business law firm headquartered in Warsaw and Krakow, with a New York office under the iSTART1 brand. Building on a practice dating back to 1981, we advise clients in the following areas.
Select a practice area to learn more.
Service directory: kg-legal.eu/services
KG LEGAL KIELTYKA GLADKOWSKI offers legal support for automotive AI projects in Poland, including cooperation between manufacturers, technology suppliers and university spin-outs. Our assistance covers technology licensing, R&D agreements, intellectual property, data protection, regulatory assessment and contractual allocation of liability.
We help clients structure arrangements for software supply, updates and ongoing support.
For advice on developing or deploying automotive AI solutions in Poland, contact office@kg-legal.pl.