Cross-border legal advice for AI developers, technology providers, investors and regulated businesses.
KIELTYKA GLADKOWSKI KG LEGAL supports businesses developing, supplying, investing in and using artificial intelligence in Poland and across borders. We bring together Polish and EU regulatory advice, sector-specific requirements, commercial contracts, data protection, cybersecurity and the allocation of responsibility throughout the AI value chain.
An AI project can connect a model developer in one jurisdiction, a cloud provider in another, a Polish integration team and customers operating across several regulated markets. Its legal position depends on what the system does, how data reaches it, who controls its deployment and what happens when its outputs influence people, products or business decisions. We help clients turn those facts into a workable legal structure for development, market entry and continuing operations.
Our work covers the lifecycle of an AI business: product design and research partnerships; access to training and operational data; investment and procurement; regulatory classification and implementation; commercialisation and licensing; and incident response, investigations and disputes. We advise on Polish and EU law and coordinate with counsel in other jurisdictions where a project requires local legal advice.
Explore our AI services:
Our AI services are designed for organisations whose legal requirements arise at different points in the technology supply chain:
We structure the engagement around the client’s actual role. A company may be a model customer in one relationship, a system provider in another and a data controller for a separate processing operation. Identifying these positions early helps determine the evidence, contractual protections and organisational measures the business needs.
Our approach builds on the interconnected framework presented in our Privacy, Data Protection, AI and Cybersecurity — Law Map. Its atomic model places data and its processing at the centre, showing how privacy, data access, cybersecurity and sector-specific rules interact within the same technological activity. Two intersecting perspectives guide the analysis.
This perspective examines the relationship between the technology and the people affected by it. We assess the collection and use of personal data, the grounds for processing, transparency, individual rights, profiling, automated decisions and international transfers. It also addresses how commercial uses of data interact with confidentiality, professional secrecy and the expectations of the people concerned.
This perspective addresses industrial and business datasets, access rights, permitted reuse, licensing, interoperability, trade secrets and the continuity of data-dependent operations. It includes the legal protection of non-personal data and the cybersecurity obligations affecting the systems through which data is collected, stored, exchanged and used.
The axes intersect. Cybersecurity protects personal as well as non-personal data, and a single dataset may contain both. A data-sharing arrangement may engage privacy rules, commercial restrictions and sector-specific requirements simultaneously. We therefore assess the rights, obligations and risks attached to each data flow, including the transition from source data to model inputs, outputs and operational records.
Effective advice starts with a clear description of the system: its inputs and outputs, intended purpose, operating conditions, dependencies and limitations. The principles used to analyse algorithms provide a useful discipline for this work. A specification explains the task; implementation determines how it is performed; testing supplies evidence; and resource constraints affect performance in practice.
We translate that technical description into legal questions that matter to a transaction, deployment or dispute:
For machine-learning systems, we distinguish implementation defects, data-quality problems, statistical performance limits and changes in the operating environment. A proof that software implements a specification and evidence that an AI model performs reliably in its intended environment answer different questions. Our legal work connects the relevant evidence to responsibilities and remedies, with input from the client’s engineers and technical specialists where required.
This analysis also helps determine whether software falls within the AI Act’s definition of an AI system. The presence of algorithms, automation or sophisticated data processing does not by itself settle that question. Similarly, evidence of technical performance must be assessed against the particular legal and contractual requirements of the intended use.
The EU AI Act, Regulation (EU) 2024/1689 as amended, provides the central framework for our AI regulatory analysis. We translate its provisions on scope, prohibited practices, high-risk systems, transparency, general-purpose models, innovation, supervision and enforcement into a practical programme for the business.
We identify the relevant EU and national supervisory bodies and assess the role of standards, codes of practice and regulatory guidance. This helps clients distinguish binding requirements from voluntary compliance tools and organise their dealings with authorities, assessors and commercial counterparties.
We assess the connection between the project and the EU market, including the location of providers and deployers, the placing of systems or models on the market, distribution arrangements and the use of AI outputs within the Union. A supplier established outside the EU may still fall within the AI Act. The analysis must account for the actual commercial and deployment arrangements, alongside any applicable exclusions.
For an international group, we help identify which entity performs each regulated activity, who holds the necessary documentation, how EU representation requirements should be addressed and how responsibilities are coordinated with overseas development and support teams.
We map the positions of providers, deployers, importers, distributors, product manufacturers and authorised representatives. We also examine whether branding, integration, modification or a change in intended purpose alters a party’s position. These AI Act roles require a separate assessment from the roles of controller and processor under data protection law.
Our work connects the resulting responsibilities to contracts: access to technical information, cooperation with assessments, disclosure of limitations, change notification, incident assistance and support for regulatory enquiries. This is particularly relevant when a customer builds a regulated application on technology supplied by several independent businesses.
We screen intended uses against the AI Act’s prohibitions and assess the applicable high-risk classification route. The review considers the system’s functionality, intended purpose, affected persons and relationship to regulated products or activities. Relevant issues may include manipulation, exploitation of vulnerability, social scoring, biometrics and certain uses in employment or public decision-making.
We document the reasoning behind the classification, including the conditions for any exception. The use of sensitive data alone does not determine high-risk status. A reliable assessment also identifies which changes would require the classification to be revisited, such as expansion into a new sector, use by a different customer group or modification of a safety-related function.
For projects subject to the high-risk regime, we help clients organise the legal work around the applicable requirements and implementation timetable. The objective is a documented allocation of tasks between management, product teams, data specialists, compliance functions and external suppliers.
We support the legal structure of risk management processes, documentation responsibilities and evidence retention. This includes defining the intended use, identifying foreseeable misuse, recording limitations and determining how training, validation and testing information will be obtained from the relevant parties.
We help align technical records with contractual statements and customer information. Where a business depends on third-party models or datasets, the agreement should address the information needed for its own assessment and the process for responding when that information changes.
We advise on the allocation of human oversight responsibilities, escalation procedures, user instructions and the authority to intervene, suspend or discontinue a system’s use. We assess whether a fundamental rights impact assessment is required and how it should interact with any data protection impact assessment. These assessments have different legal triggers, even when they can draw on overlapping evidence.
We identify the applicable conformity route and assist with the legal coordination of standards, documentation, declarations, registration and CE marking where required. The route may involve internal control, a notified body or a sector-specific product procedure. We help define the scope of external engagements and the contractual obligations of the parties supplying evidence.
We help establish procedures for monitoring deployed systems, handling complaints and serious incidents, maintaining records and implementing corrective measures. Change management receives particular attention: updates, fine-tuning, new datasets and altered uses may affect the original assessment. Responsibilities for identifying and evaluating those changes should remain clear throughout the commercial relationship.
We advise on the legal distinction between a general-purpose AI model and the applications built around it. A model provider, an application developer and a business deploying the application may face different obligations, even when they participate in the same commercial service.
For model providers, our work addresses the applicable documentation and information-sharing duties, copyright compliance policies, summaries of training content and relationships with downstream developers. Where systemic-risk requirements apply, the legal programme must be aligned with model evaluation, testing, risk mitigation, incident handling and cybersecurity processes.
For businesses using third-party models, we review the permitted uses, restrictions on customer data, rights to fine-tune, access to documentation, continuity arrangements and the consequences of model changes or withdrawal. Open-source availability requires its own assessment of licence conditions and regulatory treatment.
We help identify which transparency obligations apply to the particular system and actor. This includes notices about interaction with AI, machine-readable identification of synthetic content, disclosures concerning deepfakes and rules for certain public-interest text. We assess the relevant exceptions and transitional arrangements and help integrate the resulting disclosures into customer-facing processes.
AI agents raise practical questions about authority and control. We assist with approval requirements, permissions, spending limits, access to business systems, records of actions and escalation when an instruction is ambiguous or unsafe. Contracts and internal policies should identify who may authorise an action, when human approval is required and how responsibility is investigated if an agent sends a communication, changes a record or initiates a transaction.
We assess personal data throughout the AI lifecycle, including source datasets, training and fine-tuning materials, retrieval systems, prompts, outputs, user feedback and logs. The analysis addresses the relevant GDPR obligations and any additional confidentiality or sector-specific restrictions.
Our services include:
Anonymity requires a factual assessment. Pseudonymisation, encryption or removal of obvious identifiers may leave personal data identifiable. An AI model trained on personal data is also not automatically anonymous. We examine the relevant risks of identification and extraction in light of the European Data Protection Board’s guidance on AI models.
For international deployments, we connect privacy requirements to the actual architecture: hosting locations, support access, subprocessors, model endpoints and data retention. This enables clients to negotiate arrangements that reflect how the service operates and to address changes in providers or processing locations.
Industrial measurements, machine telemetry, business records, research datasets and operational information can be central to an AI product’s value. Legal access to those resources, and the ability to continue using them, should be assessed before development commitments or investment assumptions depend on them.
We advise on the Data Act, the Data Governance Act, database rights, trade secrets and contractual controls. The work can include access to connected-product data, business-to-business sharing arrangements, protected public-sector data, data intermediation and participation in data spaces.
We translate access into defined rights of use: who may collect, copy, combine, analyse, train on, retain, disclose or commercialise the data; for which purposes; and subject to which restrictions. We also examine what happens to derived datasets, annotations and outputs when the relationship ends.
For cloud and data-processing arrangements, we address switching, portability, interoperability, continuity and exit assistance. For confidential industrial datasets, we consider safeguards against disclosure and misuse while assessing any applicable sharing obligations.
Rights in data can arise from several sources. We therefore identify the particular access rights, contractual permissions and intellectual property protections available, together with their limits. Non-personal status does not by itself establish an unrestricted right to use a dataset, and mixed datasets require coordinated treatment of their personal and commercial elements.
We assess cybersecurity obligations against the organisation’s sector, activities and product architecture. Relevant frameworks may include the AI Act, NIS2 and Poland’s national cybersecurity legislation, the Cyber Resilience Act, DORA and data protection requirements.
Our work covers responsibility for security measures, supplier due diligence, vulnerability handling, updates, business continuity and access to records. For AI systems, this also requires attention to the legal implications of compromised training data, manipulated inputs, unauthorised access to models, confidential information appearing in outputs and failures in connected services.
We help clients determine whether their organisation or product is within a particular regime and which obligations apply to its role. A software product, a hosted AI service and a cloud supplier may require different assessments. We also examine how customer requirements become contractual obligations for suppliers even where a particular regulatory duty rests primarily with the customer.
When an incident occurs, we assist with legal triage, preservation of evidence, notification analysis, communications with counterparties and coordination with technical responders. A single event may need to be assessed under several reporting frameworks, each with its own thresholds, recipients and deadlines. An incident plan should make those decisions possible while the technical investigation is still developing.
Sector regulation is a central part of our AI practice. We assess the technology within the activity it supports, including the authorisations, professional duties, safety requirements and customer protections governing that activity.
We support the development and commercialisation of AI-enabled medical software, diagnostics, clinical decision-support tools and research applications. The work includes intended-purpose analysis, software qualification, medical-device and in vitro diagnostic requirements, clinical and performance evidence, research agreements and market-access arrangements.
We connect the AI Act assessment with MDR or IVDR obligations where applicable, and with privacy, patient confidentiality and rules governing research and clinical data. We also advise on preparation for the phased European Health Data Space framework, including access and permitted secondary use. A project plan should distinguish the rules governing current access from future EHDS arrangements.
We advise on AI used in credit processes, insurance, fraud detection, investment services, customer interactions and operational risk management. The assessment connects AI classification and data protection with the financial activity concerned, including licensing, conduct, outsourcing and supervisory requirements where relevant.
For AI and cloud suppliers serving financial institutions, we review DORA-related contractual requirements, audit and access rights, subcontracting, incident assistance, continuity and exit arrangements. For trading and investment applications, we assess the relevant financial-market rules separately from cybersecurity obligations, including responsibility for the decisions and communications the system supports.
We assist with AI used to screen applicants, evaluate performance, allocate work, monitor staff or support educational decisions. The review addresses the applicable AI Act classification, employment and equality rules, privacy, information duties and the quality of human review.
We help employers and technology suppliers define permitted functions, restrictions, evidence requirements and routes for challenging decisions. International rollouts require attention to local labour-law arrangements and consultation obligations as well as the common EU framework.
Industrial AI can connect product safety, operational continuity and valuable machine-generated data. We advise on applications in predictive maintenance, production control, quality assurance, logistics and infrastructure management.
Our work addresses product and safety qualification, allocation of responsibilities between component suppliers and manufacturers, access to operational data, cybersecurity requirements and the consequences of software updates. Contracts should explain who evaluates a change, who supplies supporting evidence and who takes action when a failure affects production or a safety-related function.
We assess AI projects at the intersection of defence procurement, sensitive information, export controls, sanctions, cybersecurity and relevant product or aviation rules. The AI Act’s treatment of exclusively military, defence or national-security uses requires careful consideration of the actual purpose and deployment.
A project with civilian and military applications needs an assessment of both pathways. We help structure development collaborations, licensing, technical access and cross-border transfers in light of the applicable restrictions, including where the same technology may be supplied to customers with different permitted uses.
We advise on conversational services, recommendation systems, synthetic media, advertising tools and customer-facing AI. Relevant issues include consumer information, commercial claims, intellectual property, personal rights, transparency and the obligations applying to the underlying platform or communications service.
For information-gathering and analytical tools, including OSINT applications, we assess source access, permitted reuse, profiling, confidentiality and the dissemination of findings. The public availability of information is one part of the analysis of its subsequent collection, combination and use.
We support suppliers and organisations evaluating AI for public-service environments. Our work addresses procurement terms, the legal basis for the activity, documentation and audit access, individual safeguards, applicable impact assessments and the allocation of responsibilities between the institution and its technology providers.
AI agreements should reflect the system being delivered and the conditions in which it will be used. We draft and negotiate development agreements, licences, model-access and API terms, data-sharing arrangements, implementation contracts, procurement documents and cooperation agreements.
Key workstreams include:
We assess the rights needed to develop and commercialise a product, including licences for training materials and software dependencies, relevant text-and-data-mining rules and reservations of rights. We also help protect confidential know-how, model-related information and business datasets through access arrangements and contractual safeguards.
For generated outputs, we assess the client’s intended exploitation, the available rights and the risk of third-party claims. Contractual allocation of rights should reflect what the applicable law can protect and what a supplier can validly grant.
In investment, acquisition and joint-venture work, we examine regulatory classification, the provenance and permitted use of data, software and model licences, dependencies on key suppliers, evidence of compliance work and exposure under existing customer commitments. Findings can be translated into transaction conditions, warranties, indemnities and a post-completion implementation plan.
Responsibility for an AI-related failure depends on the applicable legal basis and the facts of the case. The investigation may need to distinguish a software defect, inaccurate source data, a known model limitation, an unauthorised use, inadequate oversight or a failure to respond to an identified risk.
We help clients assess and manage exposure across several areas:
Our work includes analysing causation, identifying the relevant parties and preserving the records needed to establish what happened. Model versions, system configurations, prompts, logs, test results, warnings, approvals and update histories may each contribute to the evidential picture.
We advise on the transition to Directive (EU) 2024/2853 on product liability, which expressly addresses software and AI systems. The relevant regime depends on national implementation and the date on which the product was placed on the market or put into service. The new framework is due to apply to products placed on the market or put into service after 8 December 2026. Earlier products remain subject to the preceding product-liability framework.
We also review how claims are allocated between the developer, model supplier, integrator, deploying business and other parties. Contractual remedies can support that allocation, while mandatory duties and third-party rights must be assessed independently. Cross-border disputes require coordinated analysis of jurisdiction, applicable law, evidence, insurance and enforcement.
We help foreign businesses establish a legally coherent route to the Polish and EU markets and assist Polish businesses entering international technology relationships. The work combines regulatory analysis with the commercial arrangements through which the product or service reaches its users.
Depending on the project, this includes:
We prepare a division of responsibilities that can be used by headquarters, the Polish entity, overseas suppliers and local operating teams. Where several jurisdictions are involved, we coordinate the Polish and EU analysis with the advice required in those jurisdictions, so that contractual assumptions and regulatory conclusions remain consistent.
We assist management teams in establishing an AI governance structure proportionate to their business. The work can include an inventory of systems and use cases, approval processes for new deployments, procurement review, internal-use rules, responsibility matrices and escalation procedures.
We help develop AI literacy measures suited to the organisation’s personnel and activities. Training and policies can address permitted use, confidential information, checking outputs, recognising limitations, human oversight and reporting concerns. We also support the documentation of those measures and their adaptation as systems and responsibilities change.
For research and innovation, we advise on collaboration agreements, access to data, allocation of intellectual property, controlled testing and participation in regulatory sandboxes. The scope of any research exclusion or testing arrangement must be assessed against the particular activity. We help clients plan the transition from research or proof of concept to a commercial product, including the evidence and agreements required at each stage.
AI implementation plans need to distinguish obligations already applicable from those subject to transition. Following the Digital Omnibus on AI, which entered into force on 27 July 2026, the principal high-risk requirements in Chapter III, Sections 1–3 apply from 2 December 2027 for the Annex III route and 2 August 2028 for the Annex I route, subject to the relevant provisions and transitional rules.
Other AI Act requirements, including applicable rules on transparency and general-purpose models, already operate on earlier schedules. Particular arrangements apply to existing models and systems. For example, qualifying generative systems placed on the market before 2 August 2026 have a specific transition until 2 December 2026 for the machine-readable marking obligation in Article 50(2).
We build implementation plans around the client’s actual systems, launch dates and contractual commitments, alongside other applicable regimes. A later AI Act milestone does not postpone existing privacy, cybersecurity, sectoral or contractual duties. The plan identifies responsible teams, evidence dependencies and the decisions needed before each deployment or significant update.
An engagement can address a defined issue or cover a broader development and commercialisation programme. We agree the scope against the client’s product, markets, sector and stage of development.
Depending on the engagement, deliverables may include an AI classification opinion, a map of regulatory responsibilities, a data-access and transfer assessment, a contract package, a compliance implementation plan, a transaction due-diligence report or an incident and dispute-response strategy.
Whether you are developing an AI product, deploying a third-party model, entering the EU market, investing in an AI business or responding to a regulatory issue, KIELTYKA GLADKOWSKI KG LEGAL can help define the legal work required.
For an initial discussion, tell us what the technology does, where it will be offered or used, which sector it serves and what decision or transaction you are preparing. We can then identify the relevant legal questions and propose a scope of support.
Contact: office@kg-legal.pl
The relevant framework is determined for each project. Principal reference points for this practice include:
Technical background: Zbigniew J. Czech, Wprowadzenie do analizy i projektowania algorytmów, Wydawnictwo Naukowe PWN — a reference for the underlying concepts of algorithm specification, correctness, complexity and data structures.
Regulatory reference date: 12 November 2026.