Our Expert in Romania
Who this guide is for: in‑house counsel, patent prosecutors, founders and litigators in Romania who need to know whether AI‑assisted inventions can be patented, who counts as an inventor, and how to litigate inventorship disputes and enforce patents. The focus is practical: OSIM examination, Romanian evidence rules, and courtroom tactics.
AI inventorship Romania sits at the intersection of two fast-moving fields: rapid adoption of generative and machine-learning tools in research and drafting, and a patent system built around human authorship. As 2026 arrives, practitioners across Bucharest and beyond are grappling with a deceptively simple question, when an algorithm materially contributes to a technical solution, who is the inventor, and can the resulting patent survive scrutiny? This guide sets out the Romanian legal framework, examines how the State Office for Inventions and Trademarks (OSIM) is likely to treat AI-assisted inventions, and delivers a litigation-focused roadmap for proving human inventorship and enforcing patents where AI played a role.
Because Romanian courts and OSIM have limited domestic precedent, the practical standard is still being written, and careful drafting and evidence preservation now will determine who prevails later.
Understanding AI inventorship Romania begins with the domestic statutory architecture and the international instruments that shape it. Romanian patent law does not exist in isolation: it is deeply harmonised with European practice, and OSIM examiners routinely draw on European Patent Office reasoning when domestic guidance is silent. This layered structure matters because the question of whether an AI system can be named as an inventor has been answered emphatically abroad, and Romanian authorities are likely to follow that reasoning until domestic case law says otherwise.
Patent protection in Romania is administered by OSIM, the national authority responsible for examining applications, granting patents and maintaining the industrial property registers. The governing statutory framework, principally Romania’s Patent Law (Law no. 64/1991 on patents for invention, as republished and subsequently amended) together with its implementing regulations, sets out the requirements for patentability, the designation of inventors, and the rules on ownership and assignment. The consolidated legislation is accessible through the official legislative portal at legislatie. just. ro. OSIM’s examination practice, published through its official channels, translates those statutory requirements into procedural expectations for applicants.
Crucially, the statutory concept of an inventor is framed around a person who makes a creative contribution to the technical solution, and OSIM’s filing procedures require the identification of that person. For any AI-assisted invention, the application must therefore name a natural person as inventor; there is no mechanism in current OSIM practice to record a machine in that role.
Romania is a contracting state to the European Patent Convention, and the EPC legal texts exert strong gravitational pull on domestic examination. Under EPO practice, an inventor must be an identifiable natural person, and inventor designation forms are structured accordingly. When the EPO refused applications that named an AI system as sole inventor (the DABUS applications), it confirmed that the legal framework contemplates only human inventors. Romanian examiners, lacking domestic authority on the point, are likely to track this reasoning closely. Romania’s participation in the international filing system administered by the World Intellectual Property Organization reinforces that Romanian applicants operating across borders will encounter the same human-inventor requirement in every major jurisdiction.
This alignment is not incidental: it means arguments developed at the EPO and interpretations from harmonised European practice carry persuasive weight before OSIM and Romanian courts.
Practitioner experiments during 2025 and 2026, including documented drafting exercises using large language models to generate claim sets and technical descriptions, have surfaced concrete concerns about inventorship attribution, disclosure and enforceability. These experiments (presented as illustrative professional observations rather than legal rulings) show that AI can accelerate drafting but simultaneously blurs the record of who actually conceived the inventive contribution. That blurring is precisely what litigation later exploits. The takeaway for AI inventorship Romania is that the technology is already in the workflow, while the legal infrastructure to defend it in a dispute lags behind.
A recurring practical question is whether AI‑generated inventions are patentable in Romania at all, and what disclosure is required. The short answer is that an AI-assisted invention can be patented, provided it satisfies the ordinary statutory tests and the application meets disclosure and sufficiency requirements. The involvement of AI does not, by itself, defeat patentability, but it does change how carefully the application must be prepared.
Under Romanian law, harmonised with the EPC, a patentable invention must be new, must involve an inventive step and must be susceptible of industrial application. These criteria apply identically whether the technical solution emerged from a human working alone or from a human working with AI tools. Novelty is assessed against the state of the art; inventive step is judged from the standpoint of the person skilled in the art. The origin of the idea, human intuition or machine output, is legally irrelevant to these tests. What matters is the objective content of the claimed invention measured against prior art.
For AI and patents in Romania, this is reassuring: the technology used to reach the solution does not enter the novelty or inventive-step analysis directly.
The more difficult terrain concerns statutory exclusions. Consistent with EPC practice, certain subject matter is excluded from patentability as such, notably mathematical methods, computer programs as such, and abstract schemes. An AI model expressed purely as a mathematical method or an algorithm in the abstract will generally fall outside patentable subject matter. The path to protection lies in framing the invention as a computer-implemented invention that produces a further technical effect: a concrete technical improvement in a process, a device, or a system, rather than a mere algorithm.
Practitioners should therefore:
This distinction determines whether the patentability of AI inventions in Romania succeeds or founders at examination.
Sufficiency of disclosure, the requirement that the application enable a person skilled in the art to carry out the invention, takes on added weight for AI-assisted inventions. Where the inventive contribution depends on a trained model, examiners and later opponents may probe whether the specification actually enables reproduction. If the invention cannot be reproduced without proprietary training data or undisclosed model parameters, an insufficiency objection or invalidity attack becomes available. The safest course is to disclose the technical architecture, the relevant training approach at a level that enables replication of the technical effect, and the parameters necessary for the skilled person to reproduce the result.
Over-reliance on AI-generated boilerplate that has not been technically verified is a recurring risk: unverified passages can create enablement gaps that surface only during litigation.
This section addresses the central doctrinal question of AI inventorship Romania: can an AI be named as an inventor under Romanian patent law, and if not, how should applicants handle machine contributions? The answer shapes drafting, ownership and litigation strategy alike.
Under current Romanian practice, an inventor must be a natural person. The statutory framework accessible through legislatie. just. ro frames inventorship around the individual who makes the creative, intellectual contribution to the invention, and OSIM’s filing procedures require identification of that person. There is no Romanian decision that permits an AI system to be recorded as inventor, and OSIM has published no acceptance of non-human inventors. This position is reinforced by harmonised European practice: the EPO requires an identifiable natural person, and Romanian examiners are likely to apply the same reasoning. In practical terms, an application that names a machine as sole inventor will be rejected on formalities.
The correct approach is to identify the human being or beings whose intellectual contribution shaped the claimed technical solution, even where AI tools materially assisted the process.
Romanian law distinguishes the moral right of inventorship, the recognition of the person who conceived the invention, from ownership of the resulting patent. When AI assists the inventive process, the analysis focuses on the human contribution: who conceived the inventive concept, who directed the AI’s use towards the technical problem, and who recognised and refined the useful output. Where several people made distinct creative contributions, joint inventorship arises and each must be named. The AI itself is treated as a tool, analogous to sophisticated software or laboratory equipment, no more capable of being an inventor than a microscope.
The genuine difficulty is evidential: identifying and documenting the specific human contributions when much of the technical work passed through an automated system. Contemporaneous records of who framed the problem, who selected and prompted the model, and who evaluated and adapted its output become the backbone of a defensible inventorship position.
Inventorship and ownership diverge in the employment context. Where an invention is made by an employee in the course of duties, ownership may vest in or be claimed by the employer under the applicable statutory and contractual rules, subject to the employee’s right to be named as inventor and to any applicable remuneration. AI-assisted invention complicates this by making the human contribution harder to trace to a specific employee. Employers should ensure that assignment provisions capture inventions developed with AI assistance, that employees remain properly identified as inventors, and that internal records establish which employee made the creative contribution the assignment is meant to transfer.
When AI has been used in the inventive process, the inventor designation should name the natural person or persons responsible for the intellectual contribution, and internal documentation should support that designation. Prudent practice is to prepare a contemporaneous inventorship memorandum, an internal affidavit-style record, describing the human contribution, the role AI played as a tool, and the decision points at which human judgment shaped the result. This record is not filed with OSIM but becomes invaluable if inventorship is later challenged in litigation. It converts an abstract claim of human authorship into a documented, testable narrative.
How OSIM examines AI-assisted applications determines much of the prosecution strategy for AI inventorship Romania. While OSIM has not issued a dedicated body of published guidance singling out AI inventions, its established examination practice, available through OSIM’s official site, supplies the operative principles.
OSIM examiners assess AI-assisted applications against the same statutory criteria as any other application: correct inventor designation, patentable subject matter, novelty, inventive step, industrial applicability and sufficiency of disclosure. Where AI contributed to the invention, the examiner’s concern is not the involvement of the technology but the formal correctness of the inventor designation and the adequacy of disclosure. If an application appears to name a non-human inventor, or if the specification suggests the invention cannot be reproduced without undisclosed proprietary elements, examiners will raise objections. Consistent with the EPO practice that informs Romanian examination, OSIM follows human-inventor norms; there is no route to record a machine as inventor.
Applicants should therefore treat the inventor designation as a formality that must be satisfied by a natural person, and the disclosure as a substantive requirement that must enable reproduction of the technical effect.
In the ordinary course, OSIM requires the standard elements of a patent application: a full description, claims, drawings where relevant, an abstract and inventor designation. For AI-assisted inventions, examiners may probe the sufficiency of the technical disclosure where the invention depends on a trained model, seeking enough detail to enable the skilled person to reproduce the claimed technical result. Applicants should be prepared to supplement the description with the technical architecture, the functional characteristics of the model and any parameters necessary to achieve the disclosed effect, while managing legitimate confidentiality concerns through the way the invention is framed.
The core of any dispute over AI inventorship Romania is evidential: establishing, or dismantling, the claim that a named human made the inventive contribution. Romanian civil and patent procedure gives litigators a range of tools, but success turns on early evidence capture and disciplined presentation. This section sets out the burden of proof, the categories of evidence that decide these cases, and the procedural tactics that preserve and deploy that evidence.
In Romanian civil proceedings, including patent invalidation actions and ownership or inventorship disputes, the party asserting a fact generally bears the burden of proving it, in accordance with the Romanian Civil Procedure Code. A challenger seeking to invalidate a patent or to displace a named inventor must adduce evidence supporting the challenge, while the patentee defends the designation with its own record. Because inventorship disputes turn on internal facts about who conceived and developed the invention, the party with the better contemporaneous documentation typically prevails. This makes the record built during research and prosecution, not just the arguments made at trial, the decisive factor.
Where an AI system contributed to an invention, the source code of any bespoke tooling, the model weights and the provenance of training data can establish both what the machine did and where human judgment intervened. Provenance records showing who curated the data, who configured the training and who selected the model demonstrate the human contribution behind the machine output. Preserving these artefacts in their original form is essential; reconstructed or altered versions invite challenge.
System logs, file metadata, version-control histories and reliable timestamps create a chronological record of the inventive process. Commit histories can show which individual authored a given change, when a key refinement was made, and how the solution evolved from problem to reduction to practice. This timeline is often the most persuasive evidence of human conception, because it ties named individuals to specific creative decisions at identifiable moments.
Testimony from the inventors and the engineers involved supplies the human narrative that documentary evidence supports. Independent AI and machine-learning experts can explain to the court what the system was capable of, where its output required human interpretation, and why the technical result reflects human judgment rather than autonomous machine invention. Expert reports translate technical complexity into terms a court can assess and are frequently determinative in AI inventorship Romania disputes.
Digital evidence is fragile and easily challenged. Forensic collection performed to defensible standards, with a documented chain of custody from acquisition to presentation, protects the integrity of logs, code and metadata. Where evidence has been collected informally or its handling cannot be accounted for, opponents will attack its reliability. Engaging qualified forensic specialists early preserves the evidential value of the technical record.
Romanian procedure allows parties to seek interim measures and to request the preservation of evidence, which is critical where relevant code, logs or deployed systems are under the control of an opponent and could be altered or destroyed. Early applications to preserve evidence, together with requests for the appointment of a court-appointed technical expert to examine technical materials, frame the dispute on favourable evidential ground. Securing these orders before an opponent can spoliate the record is often the single most important tactical step in an AI-assisted patent dispute.
Sample pleading structure and evidence checklist
Enforcing AI inventions raises the practical question of how a patentee stops infringement and recovers loss when the accused product embeds machine-learning components. The remedies available in Romania are the standard suite, but the evidence required to obtain them is distinctive.
A patentee enforcing rights in Romania may seek injunctive relief to restrain continuing infringement and damages to compensate for loss. In assessing damages, Romanian law, in line with the EU Enforcement Directive (2004/48/EC) as transposed into Romanian law, allows the court to take account of factors such as the infringer’s unfair profits and, in appropriate cases, a reasonable royalty. Interim injunctions can be pursued at the outset of proceedings to halt infringement pending final judgment. As with any patent action, the strength of the enforcement position depends on the validity of the patent and the clarity with which the claims map onto the accused system.
Infringement analysis requires showing that the accused system practises every element of a claim. Where the accused product is an AI system, this may require access to the deployed model, its configuration or its behaviour, which the infringer controls. Litigators should anticipate this by seeking evidence-preservation and disclosure of the relevant technical materials, and by using expert analysis of the deployed system’s outputs and architecture to demonstrate that the claimed features are present. Reverse-engineering the accused system’s technical behaviour is often the practical route to proving infringement of an AI-assisted patent.
Where infringing activity or relevant evidence sits in another jurisdiction, patentees must navigate cross-border evidence gathering and coordinate enforcement across systems. Within the EU, established instruments on the taking of evidence and on the recognition of judgments assist this process. Early planning for the lawful transfer of digital evidence and for parallel proceedings preserves both the claim and its evidential foundation.
Because domestic authority is limited, comparative context helps predict how AI inventorship Romania will be resolved. The following table summarises the position across key jurisdictions and its practical implications for Romanian practice.
| Issue / Jurisdiction | Can AI be named as inventor? | Precedent / approach | Practical implication for Romania |
|---|---|---|---|
| Romania | No, inventorship requires a human under current practice (interpretation) | No Romanian case directly allowing AI inventors; OSIM follows human-inventor norms influenced by EPO practice | Name a human inventor; document the human contribution; be ready to litigate the authorship evidence |
| EPO / EPC practice | No, the EPO requires natural persons | EPO Boards of Appeal (DABUS applications) held that an inventor must be a natural person | Romanian practice is likely to track EPO reasoning on human inventorship and ownership |
| UK (DABUS litigation) | Not permitted | The UK Supreme Court held that an AI cannot be named as inventor under the Patents Act 1977 | Persuasive precedent showing courts reject non-human inventors |
| US | No for designation, a human inventor is required | US courts and USPTO guidance require a human inventor; AI-assisted inventions may be patentable where a human made a significant contribution | Useful for comparative argument and for coordinating international filings |
The following practical measures reduce risk across prosecution and ownership for AI-assisted inventions.
At a high level, employment and assignment agreements should confirm that inventions developed with AI assistance fall within the scope of the assignment, require employees to disclose their creative contribution and cooperate in documenting inventorship, and preserve the employee’s right to be named as inventor. Clauses should also require employees to maintain and hand over the technical records, logs, code and data provenance, needed to prove human inventorship if a dispute later arises.
The position on AI inventorship Romania is, for now, clear in principle and demanding in practice: a machine cannot be named as inventor, an AI-assisted invention can be patented if it meets the ordinary tests, and disputes will be won or lost on the quality of the contemporaneous record of human contribution. Because domestic case law is sparse, the standard is being set in real time, and the parties who document inventorship, draft claims around technical effect, and preserve evidence forensically will hold the advantage when litigation arrives. Anyone filing or enforcing patents that involved AI in Romania should build these safeguards into their workflow now and take specialist advice on inventorship, disclosure and litigation strategy before problems crystallise.
This article is general information and does not constitute legal advice. Specific situations should be assessed with qualified Romanian counsel.
This article was produced by Global Law Experts. For specialist advice on this topic, contact Raluca Vasilescu at Cabinet M. Oproiu, a member of the Global Law Experts network.
posted 1 minute ago
posted 19 minutes ago
posted 23 minutes ago
posted 1 hour ago
posted 2 hours ago
posted 2 hours ago
posted 3 hours ago
posted 3 hours ago
posted 4 hours ago
posted 4 hours ago
posted 4 hours ago
Find the right Legal Expert for your business
Send welcome message
