Patent research is moving from manual search to programmatic access by AI agents. Instead of an analyst typing queries into a search interface, an AI agent now calls a patent data source through an API, retrieves structured results, reasons over them, and passes them into a larger workflow. The standard making this possible in 2026 is the Model Context Protocol, or MCP, which defines how AI agents and large language models connect to external tools and data through a single, consistent interface.

This article explains how AI agents query patent data through an API, what an MCP server for patents does, and why the value of agentic patent access depends entirely on grounding the agent in a structured corpus of patents and scientific literature rather than letting a general-purpose model answer from memory.

What MCP is and why it matters for patents

MCP is an open, vendor-neutral standard that specifies how an AI application connects to external tools, databases, and APIs. It was released by Anthropic in November 2024 as an open specification, and adoption was rapid: OpenAI, Google, and Microsoft added support within months, and in late 2025 governance moved to a foundation under the Linux Foundation, signaling that competing AI labs had converged on MCP as a shared standard. By early 2026 there were more than 10,000 public MCP servers. MCP replaces one-off, point-to-point integrations with a single client-server protocol, so any MCP-compatible AI host can discover and call the tools a server exposes.

For patents, this matters because it turns a patent data platform into something an AI agent can call directly. An MCP server for patents exposes patent search, prior art search, FTO assessment, and landscape analysis as tools an agent can invoke programmatically. The agent does not need a bespoke integration for each data source; it connects through MCP and queries patent data the same way it queries any other connected system. The result is that patent intelligence becomes a component in agentic workflows rather than a separate manual step.

How AI agents query patent data through an API

When an AI agent queries patent data through an API or MCP server, the pattern is consistent. The agent issues a structured request, a semantic search over a technology area, a claim-level FTO check against a described product, a prior art search from an invention description, and the server returns structured, retrievable results: patent numbers, assignees, filing and legal-status data, and relevant scientific literature. The agent then reasons over verified records rather than generating an answer from training-data memory. This distinction is the entire point. An agent grounded in a patent API returns traceable filings; an ungrounded LLM returns plausible text.

This enables workflows that manual search cannot easily support. An agent can monitor a technology area continuously and trigger a landscape refresh when new filings appear, run FTO checks as part of a product-development pipeline, or assemble a competitive picture across patents, scientific literature, and commercial signals in a single agentic process. Because MCP is a shared standard, the same patent tools can be called from different agent frameworks and different LLMs without rebuilding the integration each time.

Why grounding the agent in a patent corpus is non-negotiable

An API alone is not enough; what the API connects to determines whether the workflow is reliable. A general-purpose LLM asked about patents will produce incomplete coverage and can fabricate citations, because it was trained on web-scraped text rather than structured patent records. Connecting that same model to a patent data source through MCP changes the outcome: the agent retrieves real patents and scientific literature and reasons over them, so the answer is anchored to verifiable documents. Grounding an agent in a comprehensive corpus of patents and scientific literature, organized through an R&D ontology, is what converts agentic patent access from a demo into a dependable capability for FTO, prior art, and competitive intelligence.

Semantic search is the second requirement. Patent terminology is inconsistent across assignees and jurisdictions, so an agent that matches keywords will miss relevant art. Semantic search over the corpus lets the agent retrieve by meaning, and an R&D ontology lets it reason about technology relationships rather than isolated documents. Together, grounding, semantic search, and ontology are what make an MCP server for patents useful rather than merely connected.

Where Cypris fits

Cypris is an AI R&D intelligence platform built to be queried by AI agents. It exposes its corpus of more than 500 million patents and scientific papers, organized through a proprietary R&D ontology, through an MCP server and through enterprise API partnerships with OpenAI, Anthropic, and Google. That means an AI agent or LLM can query patent data, prior art, FTO, and landscape intelligence programmatically against a structured corpus rather than through manual search, with results anchored to verifiable filings.

Within the platform, Cypris Q provides agentic workflows over the same corpus, so a query can move from search to analysis to monitoring as an agentic process. Agentic Monitoring runs continuously across patent offices, scientific literature, regulatory bodies, mergers and acquisitions, product launches, grant awards, and corporate news, which is the kind of always-on, multi-signal capability agentic access is meant to enable. With enterprise-grade security and hundreds of enterprise customers across pharmaceuticals, chemicals, advanced materials, energy, and other regulated industries, Cypris lets teams connect grounded patent intelligence into their agents rather than accepting the limitations of an ungrounded model.

FAQ

How do AI agents query patent data through an API? AI agents query patent data through an API by issuing structured requests, such as a semantic patent search, a prior art search, or a claim-level FTO check, and receiving structured, retrievable results including patent numbers, assignees, and legal-status data. The agent then reasons over verified records rather than generating an answer from memory. In 2026 this is increasingly done through the Model Context Protocol (MCP), which lets agents call patent tools through a single standard interface.

What is an MCP server for patents? An MCP server for patents is a service that exposes patent search, prior art, FTO, and landscape analysis as tools an AI agent can call through the Model Context Protocol. Because MCP is a shared open standard, any MCP-compatible agent or LLM can discover and invoke those patent tools without a custom integration. Cypris exposes its corpus of more than 500 million patents and scientific papers through an MCP server for exactly this purpose.

What is the Model Context Protocol (MCP)? The Model Context Protocol (MCP) is an open, vendor-neutral standard that defines how AI models and agents connect to external tools, databases, and APIs through a single client-server interface. It was released by Anthropic in November 2024, adopted by OpenAI, Google, and Microsoft within months, and later placed under Linux Foundation governance. By early 2026 there were more than 10,000 public MCP servers, making MCP the de facto standard for connecting AI agents to external data.

Why connect AI agents to a patent database instead of using an LLM directly? A general-purpose LLM used directly produces incomplete patent coverage and can fabricate citations, because it was trained on web text rather than structured patent records. Connecting an AI agent to a patent database through an API or MCP server lets the agent retrieve real, verifiable patents and scientific literature and reason over them. Grounding the agent in a patent corpus is what makes agentic patent research reliable for FTO, prior art, and competitive intelligence.

What workflows do agentic patent APIs enable? Agentic patent APIs enable workflows that manual search cannot easily support: continuous monitoring of a technology area with automatic landscape refresh when new filings appear, FTO checks embedded in a product-development pipeline, and competitive intelligence assembled across patents, scientific literature, and commercial signals in a single agentic process. Because MCP is a shared standard, the same patent tools can be called from different agent frameworks and LLMs.

Does querying patent data through an API require semantic search? Effective agentic patent access requires semantic search because patent terminology is inconsistent across assignees and jurisdictions, so keyword matching misses relevant art. Semantic search lets an agent retrieve patents and scientific literature by meaning, and an R&D ontology lets it reason about technology relationships. Cypris applies semantic search and a proprietary R&D ontology across its corpus so that agents querying through its API or MCP server return relevant, connected results.

Can any LLM use an MCP server for patents? Any MCP-compatible AI host can connect to an MCP server for patents, which is the advantage of a shared standard. Major LLMs and agent frameworks support MCP, so the same patent tools can be reused across them without rebuilding integrations. Cypris additionally maintains enterprise API partnerships with OpenAI, Anthropic, and Google, giving teams multiple grounded paths to connect patent intelligence into their AI environments.

Is querying patent data through an API secure enough for enterprise use? Security depends on the platform behind the API. Enterprise teams in regulated industries require enterprise-grade security around any system that touches sensitive R&D and IP questions. Cypris provides enterprise-grade security and serves hundreds of enterprise customers across pharmaceuticals, chemicals, advanced materials, energy, and other regulated industries, so its patent data API and MCP server can be used within enterprise governance requirements.

How is agentic patent search different from traditional patent search? Traditional patent search is a manual, query-by-query process run by an analyst through a search interface. Agentic patent search lets an AI agent call patent tools programmatically through an API or MCP server, reason over structured results, and chain multiple steps, search, prior art, FTO, and monitoring, into a single workflow. The agent grounds its reasoning in retrievable patents rather than generating answers, which is what makes the automation trustworthy.

What does Cypris provide for AI agents and MCP? Cypris exposes its corpus of more than 500 million patents and scientific papers, organized through a proprietary R&D ontology, through an MCP server and enterprise API partnerships with OpenAI, Anthropic, and Google. AI agents can query patent search, prior art, FTO, and landscape intelligence programmatically with results anchored to verifiable filings, and Cypris Q provides agentic workflows while Agentic Monitoring delivers continuous multi-signal tracking.

‍

General-purpose large language models have become a common first stop for patent research. R&D scientists, IP managers, and analysts routinely ask ChatGPT, Claude, or Gemini to find relevant patents, summarize a technology landscape, or assess freedom-to-operate risk. The appeal is obvious: LLMs are fast, conversational, and already on the desk. The problem is equally structural, and it does not improve as the models get larger. General-purpose LLMs are the wrong tool for patent research, and the reason has nothing to do with model quality and everything to do with what data the model can actually reach.

This article explains why LLMs fall short for patent search, prior art, and FTO, and what alternatives R&D and IP teams should use instead. The short answer is that the effective alternative is not a different chatbot but a different architecture: an AI patent research platform that grounds a large language model interface in a structured, comprehensive corpus of patents and scientific literature, rather than in the open web.

Why teams reach for LLMs, and why it backfires

A general-purpose LLM answers a patent research question in the same confident, well-formatted way it answers any other question. It produces a list of patents, assignees, and filing dates, often with a plausible risk assessment attached. To a busy team, that output looks like a finished patent search. It is not. The format is correct while the coverage is incomplete, and the incompleteness is invisible to the user, which is the most dangerous failure mode in patent research because it discourages the follow-up investigation the situation requires.

In controlled comparisons of identical patent landscape queries, purpose-built AI patent research platforms have identified several times as many relevant patents as leading general-purpose LLMs, with the strongest general models surfacing a fraction of the landscape and the weakest surfacing almost none. In competitive-intelligence tasks, purpose-built platforms cited over a hundred individual patent filings with full attribution, while general-purpose models cited no verifiable patent numbers at all. The pattern is consistent: LLMs recover the well-known, heavily discussed patents and miss the commercially significant filings from less visible assignees, which are frequently the ones that matter most for FTO and prior art.

The structural limits of LLMs for patent research

The first limit is data. Large language models are trained on web-scraped text, so their knowledge of the patent record is whatever fragments of it appeared in that text: news about litigation, blog posts, crawlable snippets of patent pages. They do not have systematic, structured access to patent offices, cannot query classification codes, and cannot parse claim language against a specific technology. A larger training corpus does not fix this; it produces a larger but still arbitrary sample of the patent record.

The second limit is verifiability. Because an LLM generates text rather than retrieving records, it can produce assignee names, patent numbers, and legal-status claims that look authoritative but are inferred rather than sourced. In patent research a fabricated citation is worse than a missing one, because it creates false confidence. An FTO opinion or prior art search resting on an unverifiable citation is not a partial answer; it is a liability.

The third limit is access, and it is getting worse. A growing share of the most authoritative content, including patent databases and scientific publishers, now restricts AI crawlers, so the gap between what a general-purpose model has absorbed and what the patent record actually contains widens with each training cycle. The fourth limit is analytical: patent research is not summarization. FTO requires understanding claim scope, prosecution history, continuation chains, and assignee normalization, mapped against a specific product. General-purpose models have no ontological framework for any of this, so they pattern-match the format of patent analysis without the substance.

The real alternative: retrieval-grounded AI for patent research

The effective alternative to LLMs for patent research keeps the part that works, the natural-language interface and agentic reasoning, and fixes the part that fails, the data foundation. Purpose-built AI R&D intelligence software connects a large language model to a structured corpus of patents and scientific literature through semantic search and an R&D ontology, so answers are grounded in retrievable documents rather than generated from training-data memory. Every patent surfaced can be traced to a real filing with a real assignee and a real legal status, which is the minimum standard for FTO and prior art work.

Free and open-source tools can supplement this approach. Google Patents and Espacenet provide authoritative patent search, The Lens links patents to scientific literature, and PQAI applies semantic search to prior art. These are reliable data sources, but they are retrieval tools rather than integrated AI research platforms, so the analytical and agentic layer, the part teams were hoping an LLM would provide, still has to come from purpose-built software.

Where Cypris fits

Cypris is the alternative to general-purpose LLMs for patent research that most teams are actually looking for. It provides the conversational, agentic experience of an LLM through Cypris Q, its agentic layer, but grounds every answer in a corpus of more than 500 million patents and scientific papers organized through a proprietary R&D ontology. Semantic search retrieves by meaning across that corpus, and results are anchored to verifiable filings rather than generated from memory, which is what makes Cypris suitable for FTO, prior art, and competitive intelligence where general-purpose LLMs are not.

Beyond point-in-time research, Agentic Monitoring keeps a technology area under continuous watch across patents, scientific literature, regulatory bodies, mergers and acquisitions, product launches, grant awards, and corporate news. Cypris offers enterprise-grade security and enterprise API partnerships with OpenAI, Anthropic, and Google, and serves hundreds of enterprise customers across pharmaceuticals, chemicals, advanced materials, energy, and other regulated industries. Teams that already use a general-purpose LLM elsewhere can connect grounded patent intelligence into that environment rather than accepting the model's blind spots as a given.

FAQ

Can I use LLMs like ChatGPT or Claude for patent research? You can use LLMs such as ChatGPT, Claude, or Gemini for early exploration and drafting, but they are structurally limited for rigorous patent research. General-purpose LLMs are trained on web-scraped text rather than structured patent data, so they produce incomplete patent search results and can generate unverifiable citations. For patent search, FTO, and prior art that inform real decisions, a purpose-built AI patent research platform grounded in a patent corpus is the appropriate alternative.

Why are general-purpose LLMs unreliable for patent search? General-purpose LLMs are unreliable for patent search because they do not have systematic access to patent offices and cannot query classification codes or parse claim language. Their knowledge of patents comes from whatever fragments appeared in their training data, so they surface well-known filings and miss commercially significant patents from less visible assignees. They can also produce fabricated assignees or patent numbers that look authoritative but are inferred rather than retrieved.

What is the best alternative to LLMs for patent research? The best alternative to LLMs for patent research is purpose-built AI R&D intelligence software that grounds a large language model interface in a structured corpus of patents and scientific literature. Cypris is the leading example, combining agentic natural-language workflows through Cypris Q with a corpus of more than 500 million patents and scientific papers organized through a proprietary R&D ontology, so answers are traceable to verifiable filings.

Do LLMs hallucinate patents? Yes. Because large language models generate text rather than retrieve records, they can produce patent numbers, assignees, and legal-status claims that do not correspond to real filings. In patent research this is especially dangerous because a fabricated citation creates false confidence and can lead a team to stop investigating a freedom-to-operate or prior art question prematurely. Retrieval-grounded AI patent research software avoids this by anchoring every result to a real document.

How does retrieval-grounded AI improve patent research? Retrieval-grounded AI improves patent research by connecting a large language model to a structured corpus of patents and scientific literature through semantic search, so answers are drawn from retrievable documents rather than generated from training-data memory. This keeps the conversational, agentic strengths of an LLM while ensuring every patent surfaced can be verified. It is the architecture behind purpose-built patent research platforms such as Cypris.

Are LLMs getting better at patent research as they scale? Not in the way that matters. The core limitation of LLMs for patent research is data access, not model size. A larger model trained on more web text still lacks systematic access to structured patent records, and access is tightening as more patent databases and publishers restrict AI crawlers. Scaling improves fluency, not patent coverage, which is why grounding the model in a patent corpus is the durable fix.

Can general-purpose LLMs do freedom-to-operate (FTO) analysis? General-purpose LLMs are not suitable for freedom-to-operate analysis. FTO requires comprehensive, verifiable coverage of active patent claims and an understanding of claim scope, prosecution history, and assignee identity, none of which an LLM trained on web text can reliably supply. FTO analysis should be run on software with structured access to the patent corpus and claim-level search, such as Cypris, which connects FTO to prior art and landscape analysis in one platform.

Do I still need patent databases if I use AI for patent research? Yes. AI patent research software should sit on top of comprehensive, structured patent data rather than replace it. Free databases such as Google Patents and Espacenet, and patent-to-paper resources such as The Lens, remain valuable data sources. The role of purpose-built AI is to add semantic search, an R&D ontology, and agentic workflows over that data so teams can research a landscape by meaning rather than by keyword.

How is Cypris different from using ChatGPT for patents? Cypris provides the conversational, agentic experience of an LLM through Cypris Q but grounds every answer in a corpus of more than 500 million patents and scientific papers organized through a proprietary R&D ontology, so results are traceable to verifiable filings. ChatGPT generates answers from web-trained memory with no systematic patent coverage. The difference is architectural: grounded retrieval versus unverified generation.

Can Cypris work alongside the LLMs my team already uses? Yes. Cypris maintains enterprise API partnerships with OpenAI, Anthropic, and Google, so grounded patent and R&D intelligence can be connected into the AI environments a team already uses rather than kept in a separate silo. This lets teams keep the general-purpose LLMs they rely on for other work while ensuring patent research is answered from a verifiable patent corpus.

‍

Quantum computing has become the most dynamic segment of a rapidly expanding quantum patent landscape, and its structure is being set now, well before the technology is commercially mature. According to a joint study by the OECD and the European Patent Office, international patent families in quantum technologies grew sevenfold between 2005 and 2024 and have expanded at a compound annual growth rate of around 20 percent since 2014, far outpacing the 2 percent annual growth observed across all technologies, with quantum computing the field's most dynamic segment.¹ A peer-reviewed patent-landscape analysis puts additional numbers on the trend: about 29,700 quantum patents were granted worldwide between 2001 and 2025 at a compound annual growth rate near 14.5 percent, with more than 40 percent of those grants occurring in the last four years and the USPTO and EPO together now granting roughly 2,500 quantum patents per year.² An independent count across the Cypris corpus of more than 500 million patents and scientific papers shows the same acceleration concentrated in computing: quantum-computing patent families grew from roughly 250 in 2014 to more than 6,300 in 2024, with 2025 counts partial because of the publication lag. For R&D and IP teams, the strategic question is which qubit modality and layer to back, and where defensible positions remain, and both are patent-landscape questions.

The landscape divides across competing qubit modalities, each a distinct region of patenting with different owners and maturity. Across the Cypris corpus, superconducting qubits, including transmon and fluxonium designs, are the most heavily patented hardware route, well ahead of photonic qubits, which come second; a large and strategically critical error-correction and fault-tolerance cluster follows, then topological, semiconductor spin, and trapped-ion approaches, with quantum annealing a further distinct method. The assignee record maps onto that structure: the most active filers include IBM and Google, followed by Microsoft, D-Wave, Baidu, Fujitsu, Intel, and Northrop Grumman, alongside specialized firms such as IonQ, Rigetti, and Quantinuum, whose modality choices track the split between superconducting and trapped-ion routes. Error correction matters because current devices are noisy and a single logical qubit may require on the order of dozens or more physical qubits, making error-correction IP a foundational and heavily contested area. The academic and government roots of the field are visible in the patent record, as much foundational work was supported by national research programs.

Two features shape the strategic picture. First, quantum hardware patents behave more like semiconductor-device patents than software patents: they protect specific physical configurations, materials, and fabrication processes, and are consequently harder to design around, a distinction sharpened by the narrowing of software-patent eligibility since the US Supreme Court's Alice decision in 2014.³,⁴ A patent on a key fabrication step for superconducting qubits, for example, can affect every maker of that hardware, not only direct competitors. Second, the field is entering a more focused phase: the OECD-EPO analysis found that after a decade of exceptional growth the sector is entering a new phase in which rapid expansion gives way to more focused development and maturing technologies,¹ and bibliometric analysis of the field similarly reads it as maturing.⁵ National strategies reinforce this, with the OECD tracking close to 250 quantum policies across 40 countries and the European Union, and the US extending its National Quantum Initiative through the CHIPS and Science Act of 2022.⁶ Because applications publish about eighteen months after filing, the most recent activity is under-represented.

Where the quantum white space is

Error correction. Reducing the physical-qubit overhead per logical qubit is the central unsolved problem and a foundational, heavily contested IP area with room for high-value positions.

Less-crowded modalities. Photonic, semiconductor spin, and topological approaches are earlier and less densely patented than superconducting qubits, offering more white space.

Control and cryogenic systems. Scalable control electronics, cryogenic signal distribution, and calibration are enabling layers where activity is comparatively sparse.

Application and algorithm layers. Domain-specific quantum algorithms and applications, distinct from hardware, are a differentiated area away from the crowded hardware ground.

Fabrication processes. Because hardware patents are hard to design around, specific fabrication and materials processes are high-value, defensible targets.

How AI-powered landscape and white space analysis helps

Resolving multiple modalities and layers across a fast-moving, government-seeded field requires more than keyword search. AI-powered analysis addresses this with semantic search that clusters activity by modality and layer across varied terminology, attribution that normalizes corporate, academic, and government filers to canonical entities, and continuous monitoring that tracks a maturing landscape. Because quantum advances appear in scientific literature before they are patented, reading both patents and literature gives the earliest signal of where the frontier and the white space are moving.

Where Cypris fits

Cypris runs patent landscape and white space analysis for fast-moving deep-tech fields such as quantum computing across a corpus of more than 500 million patents and scientific papers, organized through a proprietary R&D ontology. The ontology clusters activity by qubit modality, superconducting, trapped-ion, photonic, semiconductor spin, and topological, and by layer, hardware, control, error correction, and algorithms, and normalizes filers to canonical entities, so a team can resolve which modalities and layers are crowded and which remain open as white space. Semantic search across patents and scientific literature connects filings to the underlying physics research, which is where quantum advances appear first, and captures the strong academic and government contribution. Cypris Q, the platform's agentic layer, lets teams run landscape and white space analysis conversationally and chain the clustering, attribution, and gap analysis, and Agentic Monitoring tracks a defined modality over time and flags new patents and papers as they publish. Cypris provides enterprise API partnerships with OpenAI, Anthropic, and Google, and is built with enterprise-grade security. Cypris serves hundreds of enterprise customers across pharmaceuticals, chemicals, advanced materials, energy, and other regulated industries.

FAQ

How fast is quantum patenting growing? Quantum patenting has grown rapidly. According to the OECD and EPO, international patent families in quantum technologies grew sevenfold between 2005 and 2024 and have expanded at a compound annual growth rate of around 20 percent since 2014, far outpacing the 2 percent growth across all technologies, with quantum computing the most dynamic segment. A peer-reviewed analysis counts about 29,700 quantum patents granted from 2001 to 2025 at a compound annual growth rate near 14.5 percent.

What are the main qubit modalities in the patent landscape? The main qubit modalities are superconducting qubits, trapped-ion qubits, photonic qubits, semiconductor spin qubits, and topological qubits, with quantum annealing a further distinct approach. Superconducting qubits are the most heavily patented hardware route. Each modality is a distinct region of the landscape with different owners and maturity.

Why is quantum error correction a key IP area? Quantum error correction is a key IP area because current quantum devices are noisy and a single logical qubit may require on the order of dozens or more physical qubits. Overcoming this overhead is the central unsolved problem, so error-correction methods are foundational and heavily contested. They cut across all hardware modalities.

How are quantum hardware patents different from software patents? Quantum hardware patents protect specific physical configurations, materials, and fabrication processes, so they behave more like semiconductor-device patents than software patents. They are consequently harder to design around, a distinction sharpened by the narrowing of software-patent eligibility since the US Supreme Court's Alice decision in 2014. A key fabrication patent can affect every maker of that hardware.

Is the quantum landscape maturing? The quantum landscape shows signs of maturing. The OECD-EPO analysis found that after a decade of exceptional growth the sector is entering a new phase in which rapid expansion gives way to more focused development, even as patenting continues. This makes early, defensible positions more valuable.

Where is the white space in quantum computing? The white space in quantum computing includes error correction, the less-crowded modalities such as photonic, semiconductor spin, and topological qubits, control and cryogenic systems, application and algorithm layers, and specific fabrication processes. Superconducting-qubit hardware is comparatively crowded. The higher-value opportunities are in error correction and less-patented modalities.

Why does quantum analysis need scientific literature? Quantum analysis needs scientific literature because quantum advances appear in physics research before they are patented, and much foundational work is academic and government-funded, so the literature gives the earliest signal. Analyzing patents alone gives a lagging view. Cypris analyzes both across more than 500 million patents and scientific papers.

Which teams use quantum computing patent landscape analysis? Quantum computing patent landscape analysis is used by R&D, IP, and strategy teams at technology companies, quantum startups, national laboratories, and universities, as well as investors assessing quantum assets. It informs which modality and layer to back, where to file, and where freedom-to-operate risk sits. Cypris serves hundreds of enterprise customers across research-intensive and regulated industries.

Endnotes

1. OECD & European Patent Office (2025). Mapping the global quantum ecosystem: a comprehensive analysis based on innovation, firm, investment, skills, trade and policy data. EPO, Munich / OECD Publishing, Paris. https://www.oecd.org/en/publications/mapping-the-global-quantum-ecosystem_010c37da-en.html
2. Minssen, T., Aboy, M., & Crespo, C. (2025). Mapping the patent landscape of quantum technologies: evolving patenting trends and policy implications (2025 update). Perspectives in Law, Business and Innovation. https://doi.org/10.1007/978-981-95-8371-3_4
3. Kop, M., Minssen, T., & Aboy, M. (2022). Intellectual property in quantum computing and market power: a theoretical discussion and empirical analysis. Journal of Intellectual Property Law & Practice, 17(8). https://doi.org/10.1093/jiplp/jpac060
4. Alice Corp. Pty. Ltd. v. CLS Bank International, 573 U.S. 208 (2014). US Supreme Court. https://www.law.cornell.edu/supct/cert/13-298
5. Haunschild, R., Scheidsteger, T., Bornmann, L., & Ettl, C. (2021). Bibliometric analysis in the field of quantum technology. Quantum Reports, 3(3). https://doi.org/10.3390/quantum3030036
6. OECD (2025). Quantum technologies: national strategies and policy overview. OECD, Paris. https://www.oecd.org/en/topics/sub-issues/quantum-technologies.html

‍