Dear Readers, Strategists and Game-Changers,
Hope you are doing great!
Today, let's go into AI and Intangible Assets!
A team asks an AI system for a technical solution. The system proposes several approaches. An engineer selects one, changes the design, tests it and discovers an improvement worth developing. Later, the company describes the result as "an asset created by AI".
That description may be convenient from a marketing perspective. It is also incomplete and may even be a real issue for patenting the solution...
Which asset are we discussing? The generated text? The selected design? The experimental evidence? The software implementing it? The expert's understanding of why it works? A potential patent right? The confidential method for reproducing the result?
Before deciding who created the asset, we need to identify what actually became valuable. Otherwise, an argument about people versus machines can obscure the work through which a promising output became something the organisation can use.
My central proposition is simple: AI can contribute substantially to creating valuable results. But generating a result and establishing an intangible asset are different events. The connection between them deserves management.
An output is a candidate, not a conclusion
Imagine two companies receiving a similar AI-generated design. One saves it in a folder. The other tests it against a meaningful baseline, identifies the conditions under which it works, clarifies relevant features and integrates it into a repeatable workflow.
The initial file may be almost identical. The organisational positions are not. The second company has done work that the first has not: it has converted a possibility into a capability whose use and limits are better understood.
A draft meeting agenda and a manufacturing instruction warrant different scrutiny. Match the process to the consequences of relying on the result.
I would reserve the strongest language for the things that have passed the relevant tests. "Candidate solution" is not a failure label. It is an honest description of where something sits before its usefulness, control and practical implementation have been established.
Creation has several layers
For management purposes, I would separate four layers: contribution, validation, control and deployment. Contribution concerns what the people and systems did. Validation concerns the evidence that the result serves its intended purpose. Control concerns the rights, permissions and confidentiality arrangements around it. Deployment concerns the organisation's ability to use and maintain it.
These are not four legal tests, and they do not mechanically determine ownership. They are a way to prevent different questions from being answered with one vague sentence: "We created it."
A scientist may define the problem and recognise the significance of an unexpected result. A model may propose candidates that the scientist would not otherwise have considered. A technician may discover the practical conditions required to reproduce one. A software engineer may build the implementation. Legal and commercial teams may establish which uses are permitted.
Describing all of that as either exclusively human or exclusively machine-created misses the structure of the process. The more useful record shows contributions and decisions without assuming that every contribution has the same legal consequence.
Legal authorship is not the same as causal contribution
This distinction matters because intellectual-property systems ask particular questions. In the United States, the Copyright Office's January 2025 report maintains the requirement for human authorship. It recognises that human-authored expression, creative selection, arrangement or modification may be protected in an AI-assisted work; merely providing prompts does not automatically make someone the author of the resulting expression.
Patent inventorship is a different inquiry. The USPTO's November 2025 revised guidance rescinded its earlier AI-specific guidance and emphasised the ordinary inventorship standard, with AI treated as a tool and only natural persons named as inventors. Under the EPC, the EPO's J 8/20 decision likewise rejects a machine as the designated inventor.
These are jurisdiction-specific positions, not a complete worldwide answer. They do not mean that all AI-assisted work is unprotectable, nor that the human operating a model automatically owns or invented everything it produces.
For an international project, the practical discipline is to record the actual contributions and obtain an assessment appropriate to the right and jurisdictions involved. A subscription invoice, a job title and a history of prompts do not answer every authorship or inventorship question on their own.
Value can exist without a new registered right
Now consider a generated method that does not produce a patentable invention or a separately copyrightable output. Could it still be useful? Certainly, in a suitable case. It might improve an internal process, support a service or help organise knowledge the company is permitted to use.
Confidential information may qualify for trade-secret protection when the applicable conditions are satisfied, including secrecy, commercial value because of that secrecy and reasonable protective measures. Whether a particular AI-assisted result meets those requirements must be examined; the presence of AI is neither a guarantee nor a substitute for the analysis.
A public answer that anyone can readily obtain should not be promoted as an exclusive secret merely because it has been copied into a confidential folder. Conversely, an internally validated method built from lawfully used proprietary information may deserve careful treatment even without a registration number.
This is why I would examine the whole route to value rather than ask only whether the first output can be protected. The reusable advantage may sit in the validation data, the implementation, the surrounding know-how or the ability to apply the result reliably.
A worked example: a laboratory method
Consider a hypothetical materials laboratory using AI to suggest ways to reduce defects in a fabrication process. The initial suggestions contain a mixture of useful ideas, impractical parameters and approaches the team has already tried. Researchers select a subset for testing.
After several experiments, they find a workable sequence. A technician identifies an environmental condition that determines whether the result can be reproduced. An engineer modifies the control software. The team then documents the sequence, its limitations and the evidence supporting it.
What has been created? Potentially several things, each requiring its own assessment: candidate inventive subject matter, experimental records, software, confidential process knowledge and human expertise. The AI output remains part of the provenance, but it is not an adequate description of the resulting position.
The company should also ask what it was allowed to contribute to the AI environment in the first place. Were the underlying records confidential? Did collaborator terms restrict reuse? Which system and settings were used? What commitments governed access and retention? These are questions to verify, not facts to infer from the reassuring label "enterprise AI".
Finally, the team must preserve the method in a form colleagues can use. A result that works only when one researcher remembers an undocumented adjustment is not yet the same capability as a controlled, reproducible process.
Keep a creation record, not an indiscriminate archive
For work with strategic significance, I would create a short asset-creation record. It should explain the intended use, the inputs and permissions, the meaningful human and system contributions, the validation performed, the retained materials and the person responsible for the next decision.
This does not mean storing every conversation forever. Some records may contain confidential, personal or unnecessary information. Retention should be proportionate, authorised and secure. The objective is an intelligible account of what matters, not a larger ungoverned archive.
Version information can be particularly useful when the result depends on a model, software environment or dataset that may change. The record should let a qualified colleague understand what was tested and which assumptions need to be rechecked. Reproducibility need not mean an identical random output; it can mean that the relevant performance claim can be evaluated again.
This also improves attribution. Instead of erasing people's work behind "AI-generated", the company can identify who framed, selected, modified, validated and implemented the result. Accurate recognition and accurate rights analysis both benefit from a less simplistic story.
For the laboratory example, the record might be only one page plus links to the controlled experiment files. It would identify the sequence selected for use, the model-assisted suggestions that informed it, the technician's critical adjustment and the conditions tested. An unresolved rights question would be marked unresolved rather than hidden in a general statement that the project is proprietary.
The record should support a decision: release the process, experiment further, seek protection, negotiate permission or stop. Documentation earns its place when it helps someone decide or reproduce.
The asset inside the person still matters
There is a second balance to consider. The organisation may retain a useful output while the person doing the work learns very little. Alternatively, the interaction may help someone understand a difficult concept and build expertise that supports future work.
The difference should not be assumed. A January 2026 randomised study of 52 people learning an unfamiliar programming library found lower immediate comprehension in the AI-assisted group, without a statistically significant average improvement in task time. It does not establish long-term cognitive deterioration or a universal effect across scientific work.
The management implication I draw is narrower: when a workflow is also meant to develop expertise, evaluate the learning as well as the deliverable. Can the engineer explain the mechanism? Recognise an invalid result? Adapt the method when the conditions change? Know when to request specialist help?
People are not property that a company owns. Their judgement should be supported, recognised and developed. An organisation that uses AI effectively should aim to build useful shared resources while preserving the human understanding needed to challenge those resources.
Do not capitalise enthusiasm
The word "asset" also has an accounting meaning that should remain distinct from the strategic language used here. IAS 38 sets recognition and measurement criteria; research expenditure is expensed, while development expenditure qualifies for recognition only when specified conditions are met.
Generating more code, designs or documents does not automatically create more recognisable balance-sheet value. Nor should an internal estimate of saved time be presented as a financial valuation without an appropriate method and evidence.
For management, I would begin with more grounded measures. How often is the resource used? What does it improve against the relevant baseline? What does it cost to validate and maintain? Which decisions depend on it? What happens if the provider, licence or environment changes?
These questions can reveal a valuable capability without pretending that every benefit has been priced precisely. They can also reveal a growing stock of maintenance obligations disguised as productivity. Deleting unreliable or unnecessary generated material may sometimes strengthen the organisation more than producing another thousand pages.
A portfolio of resources, not a pile of files
The practical outcome is a different AI dashboard. Rather than celebrating only outputs produced, I would distinguish ideas awaiting evaluation, validated resources, approved operational capabilities and material that should be retired.
Each category implies different responsibilities. A candidate should not silently enter a safety-critical process. Validated resources need supporting evidence; approved capabilities need continued review. Retired material should not remain the easiest answer for the next retrieval system to find.
This approach also prevents a false choice between enthusiasm and prohibition. Teams can experiment quickly within a suitable environment while applying stronger controls at the point where results begin influencing important decisions. The boundary is not "AI versus no AI". It is the consequence of use.
Who builds the asset?
The answer is not always one person, and it is not adequately captured by naming the model. An intangible asset may emerge through a chain of contributions, tests, permissions and organisational choices. Different legal rights may attach to different parts, or not attach at all.
That is why I see AI-enabled asset creation as a leadership issue as much as technology and legal issues. Someone must connect the result with its purpose, evidence, rights, continuity and future development. Otherwise, the company can become extraordinarily productive at generating things it cannot confidently use.
AI can help expand the possibilities. People and organisations still have to determine which possibilities deserve reliance, investment and protection.
The objective is not to prove that humans or machines deserve all the credit.
It is to ensure that what they produce becomes something worth keeping.
