European Patent Attorney Dr. Dimitris Kouzelis has warned inventors and technology startups against using artificial intelligence chatbots to draft patent applications, cautioning that public AI tools expose secret technical data and produce legally vulnerable filings.
Writing for Greek financial news platform Capital.gr, Kouzelis explained that while generative artificial intelligence systems produce convincing text with remarkable ease, they cannot guarantee legal protection for proprietary technology. He stressed that inputting unfiled technical solutions into online chatbots risks destroying an invention's commercial value before an official application is registered.
Kouzelis identified two primary hazards for inventors using AI tools: the structural threat of confidential data leakage into public databases, and the inability of pattern matching algorithms to draft strategic legal claims for unprecedented innovations.
Data leakage and loss of patent novelty
The primary concern highlighted by Kouzelis involves the unintended disclosure of confidential technical information through public digital platforms. He noted that popular online services, including major internet search engines such as Google, Bing, and Yahoo, appear free to the public but operate by capturing user queries and submitted data.
An inventive concept often represents a startup's most critical financial asset, and its economic value relies entirely on remaining undisclosed until the official filing date. When an inventor inputs a novel technical solution into an artificial intelligence model to generate text, that solution is stored within the platform's database.
If a third party subsequently queries the platform regarding the same technical problem, the system may present the previously submitted solution. Under those circumstances, the technical information is treated as publicly known, enabling competitors to implement the technology or file a competing patent application first.
Kouzelis emphasized that the European patent system enforces an absolute novelty standard. Any technical detail made accessible to the public anywhere in the world prior to the official filing date immediately enters the state of the art, invalidating subsequent patent claims. Unlike select foreign legal jurisdictions that grant inventors a grace period after public disclosure, European patent regulations offer no grace period whatsoever.
Confidentiality risks with major tech platforms
Evaluating platform security, Kouzelis observed that relying on public artificial intelligence platforms for secrecy depends on complex user agreements, account settings, model training policies, employee access protocols, and system security infrastructure. He pointed out that very few inventors thoroughly examine terms of service before submitting proprietary technical data.
While certain professional enterprise subscriptions offer contractual confidentiality guarantees, Kouzelis warned that operational risks persist and the legal responsibility for data protection remains entirely with the inventor. Furthermore, attempting to seek legal redress or enforce intellectual property rights against major American technology conglomerates such as OpenAI, Anthropic, or Google would require prohibitive financial resources.
OpenAI, Anthropic, and Google represent leading global developers of artificial intelligence systems, producing commercial language platforms including ChatGPT, Claude, and Gemini. In contrast, a certified European Patent Attorney operates under strict professional secrecy rules established by law, guaranteeing that client disclosures remain permanently protected.
Pattern matching versus strategic patent claims
Addressing the technical limitations of machine learning, Kouzelis noted that artificial intelligence models function by identifying repeated patterns across large datasets. A patent application, however, must demonstrate novel and inventive technical features that differ significantly from prior technology and produce unexpected outcomes.
Because genuine inventions lack historical pre-existing examples in training data, artificial intelligence systems cannot draw upon established patterns to evaluate novel concepts. Kouzelis warned that this limitation creates severe vulnerabilities in patent claims, which constitute the sole legal basis for patent protection when challenged in court.
Effective patent claims must balance three conflicting legal requirements: maintaining sufficiently broad coverage to block competitor workarounds, ensuring precise clarity and technical feasibility, and remaining fully supported by the detailed description. Overly broad claims risk complete invalidation by existing prior art, whereas overly narrow claims allow competitors to bypass protection through minor modifications.
Achieving proper balance requires comprehensive knowledge of existing prior art, established court precedents, European Patent Office examination practices, and active dialogue between an attorney and an inventor. Through detailed questioning regarding potential alternative implementations, a patent attorney identifies critical technical options that must be included in the original application. AI tools do not know what they have not been told and rarely ask necessary probing questions.
Kouzelis noted that while individual inventors should avoid public tools, a European Patent Attorney may utilize specialized, closed artificial intelligence software to evaluate technical results and select the precise features required for an invention. He added that mistakes made during initial filing cannot easily be corrected later, as European patent rules strictly prohibit the addition of new technical content after the initial filing date.
AI response to patent criticism
Kouzelis concluded that while artificial intelligence serves as a helpful assistant for language refinement or reviewing public technical documents, it cannot replace a European Patent Attorney or handle confidential pre-filing disclosures. He noted that a patent is not merely a well-written document, but a legal and strategic instrument designed to withstand court litigation years in the future.
To evaluate his conclusions, Kouzelis performed a reverse engineering test by asking Anthropic's Claude AI model to critique his arguments against artificial intelligence patent drafting. Claude responded that his two core arguments represented the strongest available objections rather than legal exaggeration, noting that both possessed substantive merit.
The AI model stated that the data leakage argument is structural and difficult to counter, because an unfiled invention depends entirely on secrecy regardless of future software advances. Regarding training limitations, Claude acknowledged that while models understand general claim formatting, they lack the strategic thinking developed through direct dialogue with inventors to anticipate competitor workarounds.
Patent attorney credentials and background
Dr. Dimitris Kouzelis is a physicist holding a doctorate in mechanical engineering from Ecole Centrale Paris, a prestigious French higher education institution specializing in engineering and science. He works as a European and Greek Patent Attorney and co-founded the intellectual property firm Intellex.
Kouzelis previously served as a Director at the European Patent Office, the executive organ of the European Patent Organisation responsible for examining and granting European patents across member states. He serves on the Board of Directors of the Institute of Professional Representatives before the European Patent Office, known as the European Patent Institute.
Additionally, Kouzelis represents the Hellenic Federation of Enterprises as an advisor to BusinessEurope, the leading confederation of European business lobbying groups, and serves as President of the Intellectual Property Committee at ICC-Hellas, the Greek national committee of the International Chamber of Commerce.
