The seven deadly myths of software and AI patents

Software, artificial intelligence (AI) and machine learning are driving innovation across almost every sector, yet misconceptions about patent protection continue to discourage businesses from protecting valuable intellectual property. 

In this article, we address seven of the most common myths surrounding software and AI patents, explaining what patent offices actually look for and highlighting the key considerations for innovators seeking to protect their technology.

Myth 1: “Software cannot be patented”

Reality - The question is not “Is it software?” but “What technical problem does the software solve?”

This is probably the most common misconception. UK and European patent law exclude a “computer program as such” and the US excludes “abstract ideas”, but software-based inventions can be patented where they provide a technical contribution or technical effect. 

A technical contribution or technical effect is generally an objective improvement in a technical system or outcome. Examples of technical systems include image processing, encryption, network management, medical diagnostics, industrial control systems, sensor processing and improvements to computer operation itself.

Myth 2: “I must disclose all of my source code”

Reality - Patent specifications describe the invention, not the complete source code repository

Patent applications generally do not require disclosure of source code. Instead, the application must contain enough information for a “skilled person” to implement the invention.

For software inventions this often means describing:

• Inputs and outputs
• Processing steps
• System architecture
• Technical functionality
• Alternative implementations

The focus is on explaining what the software does and how it achieves the technical effect, rather than reproducing every line of code.

Myth 3: “Adding AI or machine learning automatically makes an invention patentable”

Reality - Using AI is not the invention. The technical result achieved by the AI is what matters

Using AI does not automatically make an invention patentable. AI systems must still solve a technical problem using a technical solution. Patent offices do not award patents simply because an invention uses AI, machine learning, neural networks or large language models. AI models and machine learning algorithms are generally considered mathematical or computational techniques.

Patentability depends on whether they provide a technical contribution or effect in the same way as usual software.

Examples of where the application of AI and machine learning models may be utilised to provide a technical contribution include:

• Medical imaging analysis
• Control of industrial processes
• Sensor data processing
• Telecommunications optimisation
• Cybersecurity and encryption
• Computer performance improvements

Myth 4: “I have to disclose my entire AI training dataset”

Reality - it’s more about the characteristics of the training data than the data itself

In most cases the actual dataset does not need to be included. However, you should generally explain:

• The source of the data
• Data characteristics
• Data quality requirements
• Data representativeness
• Training methodology
• Why the data is relevant to the technical effect

What matters is enabling a skilled person to reproduce the invention, not publishing every training example.

Myth 5: “If my idea is commercially valuable it must be patentable”

Reality - Commercial value does not automatically equal technical effect

Even if a software product or an AI model or assistant is of particular commercial value, it does not necessarily mean it is patentable. That is, a business benefit alone is normally insufficient for obtaining patent protection.

If the improvements in a process relate solely to areas such as marketing; advertising; trading; revenue generation; customer recommendations and administrative workflows, then patent protection is unlikely to be available.

Improvements in these areas are generally not regarded as technical unless there is an identifiable technical contribution beyond the business objective. Technical effects are almost always objective measures, not ones that rely on subjective measures, such as the preferences or predilections of a user.

Myth 6: “The less detail I disclose, the broader my protection will be”

Reality - A strong patent is built on detailed technical disclosure supported by specific examples

Many software and AI patents fail because they do not provide enough technical detail and are deemed insufficient. As noted above, a skilled person must be able to reproduce the invention from the patent description. Therefore, if not enough detail is disclosed, then a patent application may not be granted and provide no protection at all.

For AI models, patent offices increasingly require applicants to explain how any model is trained and what it is trained based on.

This may include details of:

• Model architectures
• Inputs and outputs
• Parameters
• Feature selection
• Training approaches
• Technical implementation details

Further, broad statements such as “an AI model determines an outcome” are frequently insufficient.

Myth 7: “Publishing or launching first and patenting later is fine”

Reality - File first, publish/launch second

Almost all jurisdictions assess your invention against everything that has been publicly disclosed prior to filing, to determine whether or not to grant your patent application. Therefore, if the core concepts or details of an invention have been publicly disclosed, patent protection may be lost in many jurisdictions. Many inventors unknowingly destroy potential patent rights by publicly disclosing their invention before filing.

Common examples can include:

• Disclosure of the concepts and research in academic conference presentations or technical papers
• GitHub repositories or online blog posts
• Product demonstrations – to outside users who are not under an NDA
• Investor pitch decks – where these are not confidential

Additionally, for many software/AI start-ups and technology companies, patent protection can be valuable before commercial launch. They can support investment, licensing, commercial partnerships, company valuation and competitive positioning.

The rules around patenting software and AI are often misunderstood, causing many businesses to overlook opportunities to secure patent protection for their innovations. By separating myth from reality, innovators can make more informed decisions about their intellectual property strategy and take the right steps to secure patent protection for commercially valuable technology before opportunities are lost.