Companies are beginning to create in a new way. Texts, images, product concepts, software code, user interfaces, technical suggestions, data products, documentation, marketing assets and invention related drafts may now emerge from the interaction between people and generative AI systems. What used to be the result of a clearly identifiable human creative or technical process may now be shaped by prompts, training data, model outputs, selection decisions, editing steps, validation, internal know how and business context. This creates a new IP management👉 Strategic and operative handling of IP to maximize value. challenge: companies must understand not only what has been created, but also how it was created, who contributed to it, who may use it, and how it can be protected or controlled.
This Deep Dive points to the upcoming OFB Fireside Chat on Protection of AI Generated Creations. The discussion will address why AI generated and AI supported outputs should not be treated only as a copyright👉 A legal protection for original works, granting creators exclusive rights. or patent👉 A legal right granting exclusive control over an invention for a limited time. law question, but as a strategic IP management issue that affects ownership, documentation, human contribution, contractual control, confidentiality, technical governance and the design of internal innovation👉 Practical application of new ideas to create value. processes.
The central challenge is that generative AI creates speed before it creates clarity. A company may be able to produce more concepts, more designs, more code, more documentation and more variations than before. But the ability to create faster does not automatically mean that the company owns, controls or can safely exploit what has been created. If the creation process remains unclear, the resulting asset may be commercially useful but legally fragile, hard to enforce, difficult to license, risky to transfer and weak as evidence.
In this sense, the protection of AI generated creations is not only about asking whether AI can be an author or an inventor👉 A person who creates new devices, methods, or processes.. For companies, the more practical question is how IP management must be organized when AI becomes part of daily creation. The answer lies in a broader control architecture that combines formal IP rights, human contribution, contracts, confidentiality, documentation, technical systems and internal rules.
AI generated creations are not one type of output
A first management mistake is to treat all AI generated creations as if they belonged to one category. In practice, companies use AI in very different contexts. A marketing team may create visual campaign material. A design team may generate interface variations. A software developer may use AI assistance for code. An R&D team may use AI to explore technical solution principles. A product team may generate user scenarios or feature concepts. A data team may create synthetic datasets, model outputs or analytical summaries. A legal or IP team may use AI to structure invention disclosures, compare documents or prepare first drafts.
Each of these outputs raises a different IP question. Some may be close to copyright. Some may be relevant for design protection. Some may contain patent relevant technical features. Some may be better protected as trade secrets. Some may be valuable mainly because they are embedded in a process, a workflow, a data structure or a customer journey. Some may not be protectable in any classical sense but may still be economically important if the company can control access, timing, quality and use.
This means that companies need a typology of AI supported outputs. They must know where AI is used, which types of results are generated, which departments rely on them, and where these results enter products, services, software, customer deliverables, technical documentation or strategic decisions. Without such a map, IP management only sees the problem after the output has already been used.
The question should therefore not begin with a generic legal assessment. It should begin with an asset question. What kind of result is being created? Does it become part of a market offering? Does it influence a technical solution? Does it contribute to customer value? Does it contain confidential information? Does it create evidence needs? Does it need to be licensed, transferred, reused or disclosed? Only then can the company decide which protection logic is suitable.
Related reading: The Deep Dive on IP in the Digital Transformation👉 Using digital technology to redesign processes, culture, and value creation now. provides the broader context for why digital products, data, software, platforms and AI related outputs require a more integrated IP approach than traditional asset categories alone can provide.
👉 https://profwurzer.com/diplex/docs/ip-and-the-digital-transformation/
The attribution problem becomes a management problem
Traditional IP systems assume that a creative or technical contribution can be traced back to identifiable people, organizations and legal relationships. Employment contracts, assignment clauses, inventor rules, copyright arrangements, service provider contracts and confidentiality obligations build on that assumption. Generative AI does not remove these structures, but it makes the factual basis more complex.
When an employee creates a product concept with the help of AI, the value may lie in the initial problem framing, the prompt sequence, the selection of one output over another, the technical modification, the combination with internal know how or the integration into a product architecture. When an external design agency uses AI tools for a customer project, the client needs to know what has actually been transferred. When a developer uses AI assistance for software code, the company needs to understand how the code was created, reviewed and integrated. When an R&D team uses AI for technical solution options, the company must distinguish between a tool generated suggestion and a human technical contribution that can be documented and evaluated.
Attribution is therefore not only a legal question. It is a management question. The company must be able to explain how a result came into existence. This matters for ownership, commercialization, licensing👉 Permission to use a right or asset granted by its owner., enforcement, funding, procurement, audits, M&A, cooperation projects and disputes. If a company cannot reconstruct the creation process, it may struggle to show that it has sufficient rights, that confidentiality was maintained, that human contribution was meaningful, or that the result can be safely transferred or exploited.
This is particularly important when AI supported outputs become part of innovation processes. AI may suggest possible technical features, alternative designs, product improvements, testing scenarios or system architectures. But the protectable or strategically relevant contribution may not lie in the raw AI output. It may lie in how employees evaluate, reject, adapt, combine and technically implement it. IP management must therefore capture the human steering of the process.
The practical implication is clear. Companies need clear internal rules for attribution. They should define when AI use must be disclosed, which AI supported results must be documented, which roles are responsible for review, and how human contribution is recorded. This does not mean that every minor prompt needs to become a legal file. It means that economically relevant outputs must not remain undocumented black boxes.
Protection requires more than formal IP rights
When companies ask how AI generated creations can be protected, the first answer often focuses on patents, copyright, designs, trademarks and trade secrets. These rights remain important. But AI supported outputs often sit in a grey zone where formal protection may be uncertain, incomplete or dependent on documented human contribution. The company must therefore think beyond the question of whether a single IP right is available.
Some AI supported results may be protectable through copyright if there is sufficient human creative contribution and the applicable legal requirements are met. Some visual outputs may support design protection if they are new, distinctive and properly attributable within the relevant legal framework. Some technical developments may become patent relevant if they involve a human technical contribution and meet the requirements of patentability. Some process knowledge, prompt strategies, model configurations, evaluation methods, training data selections or internal workflows may be protected more effectively as confidential know how or trade secrets.
Other outputs may not be strong candidates for formal protection at all. They may still be valuable because they support speed, quality, customer adaptation, simulation, testing, service delivery or internal decision making👉 The process of choosing the best option among alternatives.. In such cases, protection must come from control rather than exclusivity. Contracts, access management, technical documentation, confidentiality, version control, internal policies, data governance and speed to market may become the relevant mechanisms.
This is why the protection of AI generated creations should be treated as a layered IP management task. Formal IP rights are one layer. Confidentiality is another. Contractual allocation is another. Technical control is another. Documentation and evidence are another. Market timing and operational excellence may also matter. The stronger the business relevance of the output, the more carefully these layers need to be combined.
A company that relies only on formal IP rights may overlook important control options. A company that relies only on confidentiality may miss opportunities to secure stronger legal positions. A company that relies only on contracts may still fail if it cannot show what was actually created and by whom. The management task is to select the right combination of protection mechanisms for each output category.
Related reading: The 📑IP Management Letter on IP Strategy👉 Approach to manage, protect, and leverage IP assets. as a Functional Strategy is useful here because it explains why IP protection must be connected to the company’s strategic management👉 Strategic management is the process of planning, implementing, and evaluating. logic. AI supported outputs should be assessed not only by legal category, but by their function in the business.
👉 https://profwurzer.com/ip-strategy-is-a-functional-strategy/
Documentation becomes the evidence layer of AI supported creation
Documentation is one of the most important disciplines in the protection of AI generated creations. It connects the creation process to later ownership, protection, auditability and business use. Without documentation, an AI supported output may be useful in practice but weak as an asset. It may be difficult to prove human contribution, difficult to distinguish from generic tool output, difficult to assign to the company, difficult to license and difficult to defend.
The required level of documentation depends on the economic relevance of the output. Not every internal draft, image variation or brainstorming result needs the same level of formality. But when an AI supported result becomes part of a product, software component, technical invention, customer deliverable, design asset, dataset, training workflow or strategic business material, the company should be able to reconstruct the relevant creation steps.
Good documentation should capture the purpose of the AI use, the people involved, the approved tool environment, the relevant input materials, the treatment of confidential information, the main prompt logic, the human selection decisions, the modifications made after generation, the validation steps and the final integration into the company’s work product. The purpose is not to create bureaucracy. The purpose is to preserve evidence where evidence later matters.
This is especially important where human contribution determines whether a result can be protected or commercially controlled. If a company wants to rely on a design, a technical concept, a software component or an innovation related document, it should be able to show that human judgment was more than a button press. It should be able to explain problem framing, creative direction, technical evaluation, engineering decisions, selection among alternatives, testing and final approval.
Documentation also helps separate different forms of value. The raw AI output may be less important than the prompt structure, the internal dataset, the evaluation method, the integration into a workflow or the expert decision that turns an output into a commercially relevant result. If these elements are not documented, the company may fail to recognize where its real intangible asset👉 An intangible asset is a non-physical resource with economic value to a company. lies.
Related reading: The Deep Dive on AI in Operational IP Management explains why AI in IP work requires clear governance, human review, confidentiality safeguards and structured processes. This is directly relevant for documentation because AI supported creation must be managed as part of the organization’s operating model.
👉 https://profwurzer.com/diplex/docs/ip-and-organization/ai-in-operational-ip-management/
Human contribution must be designed into the process
Many discussions about AI generated creations focus on whether AI itself creates something. For IP management, the more useful question is how companies design human contribution into AI supported workflows. If human contribution is essential for protection, attribution, quality and business relevance, then it should not be accidental. It should be part of the process.
This starts with problem framing. A meaningful AI supported creation process begins with a business or technical problem, not with random output generation. The person or team using AI should define what problem is being addressed, what constraints apply, what internal knowledge is relevant, what quality criteria matter and how the output will be evaluated. The prompt is not only an instruction to a tool. In many contexts, it is part of the creative or technical steering of the process.
The next layer is selection. Generative AI may produce many variations, but value often arises from choosing the right direction. Selection can involve technical feasibility, customer relevance, visual coherence, compatibility with a product architecture, regulatory considerations, manufacturing constraints or strategic fit. These selection decisions may be more important than the generated alternatives themselves.
A further layer is modification. AI outputs often need editing, refinement, technical implementation, testing, coding, engineering adaptation, design iteration or integration into existing systems. This is where the company’s human expertise and internal know how become visible. A raw output becomes an organizational asset only when it is transformed into something reliable, usable and aligned with the business.
Human contribution must therefore be reflected in roles and responsibilities. Who frames the task? Who decides which AI tools may be used? Who checks the output? Who validates technical correctness? Who evaluates IP relevance? Who approves use in a product or customer deliverable? Who ensures that confidential information and third-party rights are not compromised? Without these roles, AI supported creation remains fast but uncontrolled.
Related reading: Operational IP Processes provides a useful process perspective. It shows why IP work needs repeatable, business aligned procedures before outputs can become reliable decision support and before value can be captured across functions.
👉 https://profwurzer.com/diplex/docs/operational-ip-management/operational-ip-processes/
Contracts must allocate rights before uncertainty appears
AI supported creation also changes the contractual environment. Employees, freelancers, agencies, software providers, consultants, development partners and customers may all be involved in creating or using AI supported outputs. If contracts do not address this reality, companies may later discover that important outputs are not clearly assigned, that reuse is not controlled, that confidentiality is uncertain, or that warranties do not cover the relevant risks.
For employees, the company should align AI use with existing rules on inventions👉 A novel method, process or product that is original and useful., copyright, confidentiality, software development, documentation and internal approval. Employment related rules may already allocate many rights, but they may not explain when AI use must be disclosed, which tools are approved, how human contribution is recorded, or which outputs require review before external use.
For external service providers, the issue is even more direct. If an agency, consultant or developer uses AI tools, the client company needs to know whether the provider is allowed to do so, which tools are used, whether confidential materials may be entered, whether outputs are reused for other clients, whether third party rights are respected, whether training or retention by the tool provider is excluded, and what exactly is assigned or licensed to the company.
For cooperation projects, the challenge becomes more complex. Joint development agreements, research collaborations, customer specific engineering projects and ecosystem partnerships should address AI supported results explicitly. They should clarify background knowledge, newly created results, tool use, data inputs, confidentiality, human contribution, documentation duties, publication rights, use rights, improvement rights and post project exploitation. Otherwise, AI may amplify the usual uncertainty around jointly created intangible assets.
Contracts alone are not enough. They must be connected to operational processes. A contract may say that all results are assigned to the company, but if the company cannot show what the result is, how it was created and whether the provider had the right to create it in that way, the contractual position may remain weak. This is why contractual allocation must be combined with documentation, tool governance and review.
Related reading: The Deep Dive on licensing shows why contractual rights only create value when they are embedded in governance, monitoring and post deal execution. The same logic applies to AI supported creations: allocation clauses must be operationally manageable, not merely written into an agreement.
👉 https://ipbusinessacademy.org/licensing-beyond-deals-where-value-is-really-created
Generative AI rules must be usable in innovation practice
Many companies respond to generative AI risks with policies. Policies are necessary, but they often remain too abstract. Employees need rules they can actually use in innovation practice. If the rules are too general, employees will not know what to do. If they are too restrictive, AI use may move into informal workarounds. If they are too legalistic, they may be ignored by product teams, engineers, designers and business units.
Good rules should start with tool approval. Employees need to know which AI systems may be used for which purposes. Public AI systems, enterprise environments, local models and specialized vendor tools may differ significantly in confidentiality, retention, training use, access control and auditability. A company should not treat all tools as equivalent.
The second element is input control. Employees must understand which types of information may not be entered into AI systems. Unpublished inventions, source code, customer data, trade secrets, strategic roadmaps, patent drafts, FTO assessments, licensing positions and confidential partner information may require strict limitations. It is not enough to say “do not enter confidential information” if employees do not know how internal information is classified.
The third element is output review. AI outputs should not be used blindly in products, contracts, software, patent related documents, technical documentation or customer deliverables. The company should define when human review is mandatory, which function must review, and which criteria apply. A design may need brand👉 A distinctive identity that differentiates a product, service, or entity. and legal review. Code may need technical and licensing review. Technical suggestions may need engineering and IP review. Customer material may need commercial and confidentiality review.
The fourth element is documentation. Employees should know when AI use must be recorded and what must be recorded. The documentation threshold should be linked to business relevance. The more important the output, the more important it is to capture tool use, input context, human contribution, validation and approval.
The fifth element is escalation. Employees need a simple path for unclear cases. They should know when to involve IP, legal, IT security, compliance or management. A governance system that only prohibits is less useful than a system that helps employees make better decisions.
Protection depends on the company’s digital business model
The protection of AI generated creations becomes more important when AI outputs are not isolated documents but part of a digital business model👉 A business model outlines how a company creates, delivers, and captures value.. In digital transformation, value often lies in interfaces, data flows, software layers, user experience, personalization, automation, analytics, platforms and service logic. AI supported creations may enter each of these layers.
A company may use AI to generate user interface variations, support personalized customer journeys, improve recommendation systems, create training content, produce synthetic data, refine software modules or support service workflows. The resulting value may not sit in one protectable object. It may sit in a value adding architecture that combines data, software, user interaction, process knowledge, branding and operational execution.
This changes the protection question. The company should not only ask whether a single output is protectable. It should ask where the AI supported result contributes to differentiation in the business model. Does it improve customer experience? Does it make the product more adaptive? Does it reduce operational cost? Does it create a proprietary workflow? Does it improve data quality? Does it create a stronger service layer? Does it make a competitor’s imitation more difficult?
When AI supported creations are embedded in digital business models, protection must be aligned with business architecture. Patents may protect technical implementation aspects. Copyright may protect software or creative expression where applicable. Designs may protect visual interface elements. Trade secrets may protect datasets, workflows, evaluation methods or model configurations. Contracts may control access and use. Technical measures may protect interfaces, APIs, data access and deployment environments.
The management task is to connect these protection instruments to the actual value architecture. Otherwise, the company may protect what is legally visible but fail to control what is commercially decisive.
Related reading: IP Protection of Digital Business Models is helpful because it explains how IP protection in digital settings must address customer journeys, value adding architectures, digital patents and the design of protection around business model logic.
👉 https://profwurzer.com/diplex/docs/ip-and-the-digital-transformation/ip-protection-of-digital-business-models/
From output protection to control architecture
The central shift in the protection of AI generated creations is the movement from output protection to control architecture. Classical IP thinking often starts with a finished result: a work, an invention, a design, a brand, a database or a piece of software. AI supported creation requires companies to look at the process, the inputs, the tools, the human decisions, the documentation, the contractual environment and the business use.
A single AI output may be easy to copy, hard to attribute and weakly protected. But the system that creates, selects, refines, validates and applies such outputs may be a much stronger source of advantage. A company’s prompt libraries, internal knowledge bases, approved AI workflows, expert review routines, domain specific evaluation criteria, confidential datasets, product integration logic and documentation system may together create a defensible capability.
This has an important implication for IP management. Companies should not only protect finished AI supported outputs. They should protect the creation system around them. This includes access rights, know how, internal processes, tool configurations, approved templates, review checklists, version histories, quality assurance procedures and contractual rules for external contributors.
Such a control architecture makes AI supported creation more reliable. It helps employees use AI responsibly. It gives IP teams better evidence. It helps legal teams allocate rights. It helps product teams understand what can be used. It helps management assess risk👉 The probability of adverse outcomes due to uncertainty in future events.. It also helps the company identify which AI supported outputs are worth investing in and which are merely temporary work products.
This is where protection of AI generated creations becomes part of IP and digital transformation. The issue is not only how existing IP law treats AI outputs. The issue is how companies organize value creation in an AI enabled environment. The companies that build the right control architecture will be better able to create, protect, use and commercialize AI supported results.
What companies need to do
Companies should first identify where AI supported creation already happens. This includes content, designs, software, technical concepts, data products, prototypes, customer deliverables, documentation and innovation related drafts.
- They should classify AI supported outputs by business relevance and protection logic. Not every result needs the same level of review, but economically relevant outputs must be connected to a suitable combination of formal IP rights, confidentiality, contracts, technical control and market timing.
- They should define how AI supported results are legally and economically attributed. This requires clear internal rules for employees and clear contractual rules for external service providers, agencies, consultants, developers and cooperation partners.
- They should document human contribution where it matters. Problem framing, prompt logic, selection decisions, editing, technical validation, creative refinement and final approval should be visible when the output becomes relevant for products, services, software, customer deliverables or IP protection.
- They should establish usable rules for generative AI in innovation processes. Tool approval, input restrictions, output review, documentation thresholds and escalation paths must be understandable for the people who actually use AI.
- They should connect AI supported creation to their digital business model. The relevant asset may not be the individual output, but the workflow, data structure, interface, service process, technical integration or value adding architecture in which the output is used.
- They should move from isolated output protection to an integrated control architecture. Protection of AI generated creations requires IP management, legal, IT security, R&D, product teams, business units and external partners to work within one coherent governance logic.
The protection of AI generated creations is therefore not a narrow legal subtopic. It is a practical IP management discipline for companies that create value with generative AI. The core question is no longer only whether an AI supported result can be protected. The core question is whether the company can understand, document, allocate, control and use what AI helps it create.
