Bridging the AI frontier: A guide to AI inventions in Singapore and Canada

Daniel Lee on July 30, 2026

Co-authored by Mark J. Springings – Gowling WLG

Canada and Singapore rank among the world’s leading AI research and commercialisation hubs. Canada’s strength lies in its academic ecosystem, the Vector Institute, Mila, and AMII, supported by the Pan-Canadian Artificial Intelligence Strategy. Its proximity to the United States, the world’s largest technology market, makes Canada a natural gateway for companies seeking North American patent coverage. Singapore, meanwhile, has emerged as southeast Asia’s primary gateway for AI development, driven by its National AI Strategy (NAIS 2.0) and the Intellectual Property Office of Singapore (IPOS).

For patent counsel managing cross-border portfolios, securing AI-related intellectual property in these two jurisdictions requires navigating distinct statutory frameworks, differing examination practices for Computer-Implemented Inventions (CII), and contrasting prosecution timelines.

This article provides a comparative analysis of Singapore (IPOS) and Canada (CIPO) patent landscapes, offering actionable drafting and prosecution strategies for maximising portfolio value in both jurisdictions.

AI inventorship battleground

The global debate over whether an AI system can be named as an inventor – such as Stephen Thaler’s DABUS – has met consistent resistance in both Canada and Singapore, though the underlying legal bases and administrative processes differ as the following flowchart depicts.

The Singapore position (IPOS)

Under Section 2(1) of the Singapore Patents Act, an “inventor” is defined as the “actual deviser of the invention.”

  • Legal stance: IPOS has aligned with international consensus, maintaining that the “actual deviser” must be a natural person. An AI machine lacks legal personality and cannot hold or transfer patent rights.
  • Practical practice: If an application names an AI as the inventor, IPOS will issue a registry objection. To overcome this, counsel must identify the human engineers who configured the AI’s parameters, designed the training datasets, or verified the AI’s output as the true inventors.

The Canadian position (CIPO)

Under the Canadian Patent Act (R.S.C., 1985, c. P-4) and Patent Rules (SOR/2019-251), the term “inventor” is not expressly defined as a “natural person.”

However, the requirement that an inventor be human derives from the common law understanding that ‘devising’ an invention presupposes human mental activity. CIPO’s administrative position, rather than binding judicial precedent, currently governs this area.

  • Legal stance: In the Canadian prosecution of the DABUS application (CA 3,137,029), CIPO took the position that under Canadian law, the inventor must be a human being capable of exercising the mental faculties necessary to devise an invention. Because the applicant named an AI system as the sole inventor, CIPO issued a requisition that ultimately led to the application being refused. It’s worth noting, the precise procedural outcome, whether formally ‘refused’ or ‘abandoned’ for failure to comply, should be verified against official CIPO records, as these are distinct outcomes under the Patent Rules.
  • Practical practice: Under Canadian practice, the applicant must provide the names of the inventors pursuant to Rule 36 of the Patent Rules. Unlike in the US, Canada does not require a formal inventor’s oath or declaration. The chain of title from inventor to applicant is established through assignments or employment agreements. CIPO will refuse any application where the chain of title cannot be traced back to a human inventor, as an AI cannot legally hold or assign patent rights.

Patent eligibility of AI and computer-implemented inventions (CII)

Both jurisdictions reject AI as an inventor, but they diverge sharply on the patent eligibility of AI software itself regarding machine learning models, neural network architectures, and training methods.

Singapore’s pragmatic ‘Invention-as-a-Whole‘ approach

IPOS is highly receptive to software and AI patents, utilizing a holistic framework that mirrors the European Patent Office (EPO) but is possibly applied more flexibly.

  • The test: To be patentable, an AI invention must make a ‘technical contribution’ or solve a technical problem with a technical solution.
  • Software friendliness: IPOS does not categorically exclude computer programs. If an AI algorithm is integrated into a physical process, improves the internal functioning of a computer such as faster processing, better memory allocation, processing real-world physical data, analysing medical images or seismic data, then it is considered patent eligible.
  • Drafting tip for practitioners: Focus the specification on the technical utility. Describe how the AI model optimises database queries, reduces network latency, or improves hardware efficiency.

Canada’s complex Purposive Construction and the Choueifaty framework

Canada has historically been a difficult jurisdiction for computer-implemented inventions. CIPO’s former problem-solution approach often stripped software steps out of claims. The Federal Court’s decision in Choueifaty v Attorney General of Canada, 2020 FC 837, forced CIPO to change course.

  • Post-Choueifaty test: Following the Federal Court’s decision in Choueifaty v. Attorney General of Canada, 2020 FC 837, CIPO revised its approach to computer-implemented inventions. CIPO now applies ‘purposive construction’ in accordance with the Supreme Court of Canada’s precedents (Free World Trust v Électro Santé Inc., 2000 SCC 66; Whirlpool Corp. v Camco Inc., 2000 SCC 67). The examiner identifies the ‘essential elements’ of the claims as construed by the skilled person. If the computer or AI algorithm is an essential element of the claim, the question is whether the claim as a whole defines patentable subject matter under s. 2 of the Patent Act – an ‘art, process, machine, manufacture or composition of matter’ – and does not fall within the exclusion for ‘mere scientific principles or abstract theorems’ under s. 27(8).
  • Mere Scientific Principle: Section 27(8) of the Canadian Patent Act prohibits patenting ‘mere scientific principles or abstract theorems.’ CIPO may object to AI applications by characterizing machine learning algorithms as abstract mathematical methods. However, under CIPO’s post-Choueifaty guidance, examiners assess whether the claims as a whole, when purposively construed, amount to only an abstract theorem or scientific principle. A claim that applies a mathematical method to achieve a practical result may still be patentable.
  • Drafting tip: To survive a CIPO examination, claims must tightly couple the AI algorithm with a physical system or practical application. For example, instead of claiming ‘a machine learning model that classifies data,’ claim ‘a computer-implemented method for controlling an industrial manufacturing robot using a neural network trained on sensor data.’

Head-to-head comparison

The table below outlines the critical operational and legal differences patent counsel must consider when coordinating filings in Singapore and Canada.

Feature / metricSingapore (IPOS)Canada (CIPO)
Primary statutory hurdleTechnical contribution requirement (CII Guidelines).Section 2 (definition of “invention”) and Section 27(8) Patent Act (exclusion of abstract theorems).
Inventorship standard“Actual deviser” (Natural person only).Natural person only (DABUS application rejected).
Drafting focusTechnical effect and solving a technical problem.Purposive construction; coupling software with physical systems.
Prosecution speedPossibly fast: SG IP FAST program can grant patents in as little as 6 months.Moderate to slow: Standard prosecution typically takes 2–5 years, depending on technology area.
Acceleration optionsSG IP FAST, ASPIRE, Patent Prosecution Highway (PPH).Patent Prosecution Highway (PPH), Advance Examination.
Software patent friendlinessHigh: Highly predictable and software-friendly examination guidelines.Moderate: There is scrutiny of abstract algorithms and requires updated case law.
Grace period12-month public disclosure grace period.12-month grace period for disclosures made directly or indirectly by the applicant (or someone who obtained information from the applicant) before the filing date.

Strategic prosecution playbook for cross-border portfolios

For counsel managing an AI-driven patent portfolio spanning both regions, they may wish to consider the following filing strategy once the patent specification has been drafted, since Singapore and Canada participate in a bilateral PPH agreement.

  1. Consider filing in Singapore and expediting prosecution proceedings.
  2. Leverage the PPH to accelerate a corresponding Canadian application (this approach also provides a more predictable outcome).

To satisfy both IPOS’s ‘technical contribution’ standard and CIPO’s ‘purposive construction’ rules, use a tiered claim strategy in applications.

  • Base layer (for Canada): Ensure there are claims that explicitly integrate the AI model with physical hardware, specific inputs – raw sensor data, real-world images, and tangible outputs such as mechanical actions, concrete physical parameter adjustments.
  • Optimisation layer (for Singapore): Include claims focused on the algorithmic efficiency of the AI itself – unique neural network layers, weight adjustment techniques, or data pipeline structures – to capture broad protection for the core software IP.

For Canadian tech companies tapping Singapore’s talent pool, and vice versa, employment contracts and consulting agreements should explicitly define ownership of AI-assisted outputs and include pre-signed assignment clauses. This keeps the human-inventor assignment chain legally airtight from day one.

Companies with US market ambitions should also consider Canada’s role as a North American gateway. For instance, a Canadian patent, combined with a US filing, provides integrated continental coverage, and prosecution work product from CIPO can be leveraged under the Canada-US PPH to accelerate USPTO examination.

Conclusion

Canada and Singapore both require human inventorship, but they diverge sharply on how they evaluate the patentability of AI software. Singapore offers a fast, predictable pathway for securing software patents; its SG IP FAST programme can deliver a grant in as little as six months, making it an ideal launchpad for global AI portfolios.

Singapore’s software-friendly examination guidelines and its role as a gateway to ASEAN markets add further strategic value for companies expanding across Asia-Pacific. Canada, with its significant AI market and direct access to the United States, serves as a complementary gateway for North American patent coverage. The shared border, integrated supply chains, and USMCA framework make Canadian patent protection a natural complement to any US filing strategy.

That said, Canada demands careful claim drafting that integrates software with practical applications, and tactical use of prosecution tools like the PPH to meet the legal requirements for patent eligibility under the Patent Act.

Together, Singapore and Canada offer a powerful combination – Singapore provides speed, predictability, and access to Southeast Asian markets, while Canada provides depth, North American reach, and a springboard into the US.

A dual-compliant drafting strategy, combined with Singapore’s fast-track system to pull Canadian applications forward via the bilateral PPH, lets patent counsel build a formidable IP position across both economies. For companies with US market ambitions, a Canadian patent adds continental coverage at lower cost than a US filing alone, while the Canada-US PPH provides a further acceleration pathway into the USPTO.

Contact the authors or a member of Gowling WLG’s Patent team if you have any questions.

Integrating enforcement-ready IP strategies into early-stage R&D

Daniel Lee on June 29, 2026

A common pitfall for innovative companies is treating Intellectual Property (IP) as an administrative afterthought or even a form filing exercise that happens only when a product is designed and ready for market. It’s a siloed approach that often results in patents that are not properly scoped, unprotected trade secrets, and missed opportunities. More critically, it can lead to IP that are impossible to enforce against competitors.

To build real enterprise value, companies must weave IP strategy directly into the fabric of their Research and Development (R&D) processes. Outline here are best practices for capturing early-stage inventions, details on how to build an integrated R&D-IP lifecycle, and useful tips on how early-stage habits lay the foundation for effective IP enforcement in the future.

The integrated R&D-IP lifecycle

To catch high-value inventions before they are lost to public disclosure or competitor pre-emption, IP checkpoints must be incorporated into the R&D process. A typical flowchart can be shown, along with the corresponding IP checkpoints, as follows:

Ideation and concept (IP landscaping)

  • The Process: Ideally, even before writing a single line of code or synthesizing any molecules or compounds, the various research teams should work with IP counsel, whether in-house or external, to conduct a preliminary patent landscape search. This gives a better understanding of what has been done and any potential gaps in the field of technology.
  • The R&D benefit: Helps prevent “reinventing the wheel” and directs R&D budgets toward white spaces — areas with high market potential and low competitor patent density.

Proof of concept and feasibility (Invention disclosure)

  • The process: As soon as a technical solution works in a lab or simulation, engineers should prepare and submit an Invention Disclosure Form (IDF) to trigger the process of evaluating the invention for potential IP protection.
  • The R&D benefit: It establishes an early, verifiable timestamp of conception, crucial for resolving priority disputes or establishing inventorship. This can result in a patent being drafted, or the securing of a trade secret, depending on factors such as importance of the invention, detectability, licensing potential, and so on.

Product development (Filing triage)

  • The process: IP counsel reviews IDFs to decide the protection mechanism: patenting, maintaining as a trade secret, or defensive publication.
  • The R&D benefit: Ensures that core technical differentiators are legally protected before beta testing or external pilot trials begin.

Pre-launch (Freedom to Operate – FTO)

  • The process: Conduct a targeted FTO search on the finalised product design to ensure it does not infringe active patents in key markets, and the jurisdiction of search can depend on the markets of interest or manufacturing source.
  • The R&D benefit: Gives the business the green light to launch, preventing catastrophic post-launch injunctions and costly design rollbacks.

Best practices for early-stage inventions

To maximise the commercial and legal value of early-stage innovations, organizations should deploy three foundational practices.

1. The trade secret versus patent triage protocol

Not every invention should be patented. Patenting requires publishing the technical details to the world, which can act as a roadmap for competitors. Companies need a clear triage matrix:

CriterionPatenting PathwayTrade Secret Pathway
DetectabilityHigh (Competitor’s infringement is easily visible in the end product).Low (Invention is a manufacturing process or back-end algorithm).
Reverse-engineeringEasy (Competitors can easily duplicate the tech once they see it).Extremely Hard (Even with the product, the secret formula cannot be deduced).
Lifespan of TechModerate to Long (Tech will remain relevant for 5–20 years).Short (Fast-evolving code) OR Ultra-Long (e.g., Coca-Cola formula).
Regulatory FilingRequired (Publicly registered with patent offices like CIPO or IPOS).Strictly forbidden (Must be protected via encryption, NDAs, and access controls).

2. Establish ‘Cleanroom’ documentation practices

Whether using traditional paper notebooks or modern electronic lab notebooks, R&D teams must maintain clean, chronological, and unalterable records.

  • Notebooks should clearly state: Who conceived the idea, when it was conceived, who assisted, and what external resources (e.g., open-source software, third-party libraries, AI engines) were used.
  • Invention disclosure forms can assist in naming the inventors and the process in which the invention was arrived at.
  • Care should be taken in identifying the initial pool of people involved in conceiving the invention as the final form of the claims may not correctly identify the inventors, as well as identifying if any subcontractors or developers were involved.

3. Bulletproof joint R&D and contractor agreements

In the early stages, companies frequently use third-party developers, academic researchers, or engineering contractors.

  • The trap: Under Canadian, Singaporean, and US patent laws, unless there is a written contract stating otherwise, the inventor initially owns the rights.
  • The fix: Ensure every employment agreement and independent contractor agreement contains an “automatic, present assignment” clause (e.g., “Contractor hereby assigns all right, title, and interest in and to all Intellectual Property…” rather than a mere promise to assign in the future like “Contractor agrees to assign…”).

How early IP habits build enforcement-ready assets

The ultimate test of an IP asset is its utility in a dispute—whether to secure an injunction, extract licensing royalties, or defend a market monopoly. Here is how early-stage R&D practices directly enable successful litigation and enforcement:

Draft claims for ‘detectability’

The biggest mistake early-stage companies make is drafting patent claims around how an invention is made rather than how it behaves in the market.

  • The problem: If you patent an optimised server-side processing algorithm, you will struggle to enforce it. You cannot prove a competitor is using it without getting a court order to inspect their private server code — a highly expensive and legally difficult threshold to meet, especially under strict Canadian and Singaporean discovery rules.
  • Best practice: Instruct your patent agent to draft claims that target externally observable indicators or detectable API outputs.
    • Example: Instead of claiming the internal database sorting steps, claim “a user interface that displays [specific data structure] within  milliseconds of receiving [specific user query],” or target the specific data packets transmitted over the network. This makes proving infringement as simple as buying the competitor’s product and running a network packet analysis.

Build a multi-layered claim pyramid to survive validity challenges

When you sue a competitor for patent infringement, their first line of defence is almost always to counter claim your patent is invalid based on the principle of prior art.

  • Best practice: During the early R&D phase, do not just file one massive, broad claim. Work with counsel to draft a highly structured claim set containing broad independent claims for capturing competitors and a tight cascade of narrow, highly technical dependent claims as fallback positions.
  • The enforcement advantage: If a competitor unearths a surprise prior art document that invalidates your broad independent claims, your narrow dependent claims, which protect specific, optimised engineering parameters developed in Stage 3 of R&D, will survive. A single surviving dependent claim is all it takes to win an infringement lawsuit.

Maintain clean chains of title for immediate standing

In major patent jurisdictions, including Singapore and Canada, a plaintiff must have clean, unbroken legal standing to sue for infringement.

  • The problem: If a startup tries to file an emergency injunction to stop a massive competitor from launching a knock-off but has a missing assignment document from a disgruntled former contractor, the court will deny the injunction.
  • Best practice: By executing present assignments during Stage 2 and Stage 3 of the R&D process, you ensure that the company holds 100% undisputed legal ownership of the IP. When a copycat emerges, your litigation team can file for an immediate interlocutory injunction without a multi-month delay spent tracking down former employees to sign paperwork.

Leverage fast track systems to build injunction weaponry

Early-stage R&D often moves faster than patent offices. A competitor might launch a copycat while your patent is still pending in a multi-year examination backlog.

  • Best practice: Use expedited examination programs to secure granted claims quickly. For example, you can fast-track applications in Singapore via SG IP FAST to get a granted patent in as little as six months.
  • The enforcement advantage: In jurisdictions like Canada or Singapore, you cannot sue or obtain an injunction on a pending application. Getting a rapid, strategically tailored grant on a specific copycat design allows you to serve a cease-and-desist letter with teeth, halting the competitor’s market momentum before they can establish market share.

Conclusion

Enforceable IP is not generated by lawyers sitting in a vacuum – it is forged on the R&D lab bench and protected through deliberate steps. By embedding IP checkpoints into the R&D lifecycle, keeping meticulous documentation, and deliberately drafting claims with “detectability” in mind, innovative companies can transform their R&D departments from cost centres into legal fortresses. An IP strategy that is built for enforcement from day one ensures that when your technology succeeds in the market, your legal rights will successfully defend it.