The number of self-filed standard patent applications has also increased, being 82% higher in the first 10 months of 2025 compared with the previous five years’ average. However, the number of standard patent filings by self-represented applicants remains a small proportion of the total – just 2.6% of all applications filed up until the end of October – and so the significance and impact of this increase remains to be seen.
The only plausible explanation I can think of for this sudden jump in filings by self-represented applicants after years of relative stability in numbers is the increasingly widespread and affordable availability of generative AI. What is not yet apparent is how applicants are using AI. Are they using ChatGPT and similar tools to assist in drafting patent specifications describing inventions made wholly by human inventors? Or are they also using AI to facilitate invention itself?
Either way, I fear that this will not end well for many of these self-filers. To be clear, there is absolutely no question that AI tools based on large language models (LLMs) can be used to assist in drafting patent specifications. In the hands of an experienced patent professional who understands the invention to be protected, the full legal requirements for protection, and the various national and international drafting principles, even a general-purpose tool such as OpenAI’s ChatGPT, Anthropic’s Claude, or Google’s Gemini can accelerate the process of preparing a quality patent specification. But there are also many potential pitfalls to using these tools, and they do not embody the significant expertise, skill and experience of a competent patent professional. And if the AI is also contributing to the invention, then there may be nothing that is legally protectable at all!
The good news for Australian patent attorneys is that the increase in DIY (with AI) provisional applications has not been accompanied by a corresponding decrease in applicants engaging professional assistance. The number of provisional applications filed via registered attorneys and firms for the first 10 months of 2025 is down by just 3.2% on the past five years’ average. And while this does reflect an ongoing decline over recent years, it indicates that the use of AI may be bringing new users to the patent system, rather than taking work from professional advisors.
Provisional Application Filings
The chart below shows the total number of provisional applications filed between January and October (inclusive) each year since 2020, as well as the breakdown between applications filed by registered attorneys/firms, self-filed applications (i.e. those for which the ‘address for service’ matches a named applicant or inventor), and others (i.e. where the ‘address for service’ is neither an identifiable attorney/firm nor a match for an applicant or inventor). As can be seen, the number of self-filed applications has been stable for a number of years, but so far in 2025 has dramatically jumped to exceed the number of attorney filings. I have looked back three decades, to 1995, and this is the only time over this period that the number of self-filed provisional applications has exceeded the number of attorney-filed applications!
There were 2,563 provisional applications filed by self-represented applicants in the first 10 months of 2025, compared with 993 in 2024, and an average of 934 over the five years from 2020 to 2024. By comparison, the number of provisional applications filed via a registered patent attorney or firm over the same period fell slightly from 2,456 in 2024 to 2,416 in 2025. This reflects a longer-term trend – the number of attorney-filed provisional applications last peaked in 2019, with 3,013 filings in the first 10 months of that year before declining over the following years, but has remained relatively stable since 2022. The average over the five years from 2020 to 2024 was 2,496.
(I am, intentionally, largely ignoring the ‘other’ category, since in these cases the relationship of the named ‘agent’ to the inventors and/or applicants – and whether or not a patent attorney may have been involved – is unclear without investigating each application individually. At least some of these, however, will also effectively be ‘self-filings’ through a related person or entity.)
As has always been the case, the self-filers are overwhelmingly Australian residents – 2,526 of the 2,563 self-filed provisional applications in 2025 name at least one Australian applicant.
Most are also new (or long absent) users of the patent system. A total of 1,121 personal applicants and 349 corporate applicants that have named on self-filed provisional applications in 2025 have not been named on any earlier application filed since 2010. Collectively, these ‘new’ personal applicants have filed 1,587 provisional applications, while ‘new’ corporate applicants have filed 557 provisional applications.
A number of these ‘new’ applicants have filed relatively large numbers of provisional applications. The top four applicants are all personal filers who have filed 48, 25, 24, and 22 applications, respectively. The top corporate applicant, at number five overall, has filed 20 applications. Based on application titles, there is a reasonable degree of coherence among the applications filed by companies. Three of the leading corporate filers, for example, appear to be targeting the ‘digital afterlife’, nutrition products, and pharmaceuticals, respectively. Most leading personal applicants, on the other hand, appear to have filed applications covering a motley assortment of subject matters, with AI, quantum, digital security, energy and environmental technologies all being popular – sometimes in combination!
Standard Application Filings
The first thing that needs to be said about standard patent application filings in the first 10 months of 2025 is that the overall numbers appear to be significantly down on recent years. Up until the end of October, the number of standard patent applications filed in Australia (both direct/convention and PCT National Phase Entry) was just 22,420, compared with 24,777 for the same period in 2024, and 26,875 in 2021 (the historical maximum). Indeed, the number of standard application filings have not been lower at the same point in the year since 2014 (21,445), which was affected by the impending commencement of the Raising the Bar IP law reforms causing many filings to have been brought forward into 2023, reducing the number of 2024 filings.
This, in itself, is not good news for the Australian patent profession. Applications filed through a registered attorney or firm fell by 11%, from 24,238 in the first 10 months of 2024 to 21,600 over the same period this year. Self-filed standard applications increased by 50%. albeit off a low base of 373 filings in 2024 to 559 in 2025. (Applications in the ‘other’ category rose from 166 in 2024 to 261 in 2025.)
Most of the self-filed standard applications do not claim priority from any earlier filing, and are therefore still some time away from their 18-month publication dates. I was unable to find any published standard patent application filed within the past year that appeared to have been drafted with the assistance of generative AI. We are just going to have to wait and see what may be revealed over the next 12 to 18 months.
Why This Is (Almost) All Bad News
An optimist might argue that this is all very positive – perhaps generative AI is actually democratising the patent system, and making the preparation of patent specifications accessible to everyone, without the need to engage expensive patent professionals! And while there may be some instances of this occurring, I believe that it will turn out not to be true in the overwhelming majority of cases.
Historically, self-represented applicants have fared very poorly in the patent system. When I looked specifically at this topic in 2017, I found that over 90% of all self-filed provisional applications result in no further action. At A$100, the provisional filing fee is low, but that is still $100 wasted if the application then goes nowhere (or A$4800, if you have filed 48 provisional applications). More than half of all self-filed standard applications (and, at the time, innovation patent applications) made no claim to any earlier priority, indicating that the applicants were unaware of the considerable financial and strategic benefits of commencing the patenting process with a provisional application. The vast majority of standard applications – nearly 80% – produced no outcome of any value whatsoever.
I don’t anticipate that AI-assisted drafting will significantly change these outcomes. For personal applicants, with their largely speculative (and, perhaps, AI-generated) inventions, and lack of realistic commercial prospects, I expect that it will continue to be the case that few of these applications progress any further through the system.
Corporate applicants, some of which appear to be filing applications targeting commercial objectives, may exhibit a higher rate of conversion to standard applications (in Australia and/or internationally) than has historically been the case. However, I think it likely that these applications will continue to run into trouble further down the track, because the applicants will most likely be unaware of the limitations of their AI-drafted patent specifications.
Limitations of Generative AI for Patent Drafting
Generative AI can turn a description of a collection of components into a plausible-looking claim directed to that same collection of components. But it has a very limited concept of an ‘invention’, and even less of what would constitute a defensible and commercially useful scope of protection. For anything that is genuinely novel and inventive, and therefore largely unprecedented in its training data, generative AI is typically incapable of taking a high-level description of what an invention does and producing a detailed description of how it can be put into practice by an uninventive person skilled in the relevant art, which is a minimum legal requirement for a valid patent specification. By contrast, an experienced patent attorney will identify what is needed and will work with the inventor to identify and document all of the required information.
Generative AI also does not understand the important legal relationship between the scope of the description and the scope of the claims, i.e. that you cannot claim broadly unless you also describe how to practise the invention across the full scope of the claims – although this is mitigated by the fact that it tends to claim narrowly, by adhering quite closely to any description it is given. It has even less ability to anticipate potential hurdles in examination, and to ensure that the description and claims include appropriate description and fallback positions to address prior art that may emerge.
In the hands of an experienced patent drafter, generative AI can certainly be guided through many of these requirements, and be a useful drafting assistant. But without expert guidance, control and input, AI is most likely to produce something that might superficially appear more professional than the typical self-drafted patent specification, but which is just as lacking in the necessary substance, because the user was simply unaware of what was actually required, and how to provide it. And, as any patent attorney knows, but many inexperienced applicants do not, anything that is not present in the application when initially filed cannot be added later – at least, not without sacrificing the original priority date.
And all this is before we even consider the ways in which using AI to generate patent specifications may, in itself, be fatal to the validity of any potential patent rights. For starters, it is vital to ensure that entering information about the invention to the AI tool does not comprise a non-confidential disclosure that could deprive the subsequently-claimed invention of novelty. A person submitting information to a free service, or a free tier of a service such as ChatGPT, is very likely to have implicitly agreed that their interactions can be used for improvement and training of the service. Paying users are more likely to at least have the option not to have their data used for improvement and training purposes, but it is important to read the terms of service carefully, and ensure that any steps that might be required to opt-out are taken.
Further, if the AI is contributing in any way to the actual substance of the invention, then it is essential to be aware that courts in most major jurisdictions have confirmed that AI cannot be an inventor – or, at the very least, not the sole inventor – for the purposes of obtaining a patent, and that any application or patent that does not have at least one legally recognised human inventor will not be valid. The laws around inventorship can be complex, and they vary between jurisdictions, making it very unlikely that this is a matter that a lay person will be able to reliably assess for themselves.
Conclusion – ‘Slopplications’ Are Here … to Stay
The reality is that, despite all of the issues I have discussed above, inexperienced and unqualified people are using, and will continue to use, generative AI to draft patent specifications. The term ‘AI slop’ exists – and is Macquarie’s Word of the Year – precisely because AI is being widely used to generate spam and other low-quality content in order to save money on paying real writers, or to directly make money by generating revenue-raising clicks. It’s largely rubbish, but it is drawing people in, wasting their time, and possibly turning all of our brains to mush.
AI generated patent slopplications are no different: superficially coherent, professional in appearance, free of typos and likely with perfect spelling and grammar, but lacking in any of the nutritional substance that is necessary to support valid patent rights with any kind of meaningful commercial value.
If the Australian Patent Office is receiving hundreds of these kinds of applications, I can only imagine what must be happening in larger markets. The USPTO is probably receiving tens of thousands! And we can only hope that most of the provisional applications do not proceed to full applications and enter examination. They will be a nightmare to examine, and their unrepresented applicants will be difficult and time-consuming for examiners to deal with. The likelihood that AI will be used to generate submissions in response to examination reports will probably make matters worse rather than better – the human applicants will literally be filing responses they don’t understand to reports that they didn’t understand. Of course, if we can replace the examiners with AI, then we can cut the humans out of the process altogether, and the entire patent system will simply drown in slop! (Kidding … maybe not kidding.)
Yet it is possible that large, experienced, sophisticated applicants will place the system under even greater strain. The potential to use AI to, for example, churn out thickets of patent specifications and claims surrounding a core invention could multiply the number of applications filed in various patent offices. It has recently been reported that one applicant in Japan, SoftBank Group, has massively increased its patent filings to around five times the second-placed applicant (Canon), and that AI is likely to be a factor in this activity. I anticipate that this is only the beginning and that, again, it is quite possibly already under way in the US on a larger scale under the cover of non-published, and yet-to-be-published, applications.
I hope that this turns out not to be as bad as I fear, but I fear it may turn out to be worse than I hope!
Before You Go…
Thank you for reading this article to the end – I hope you enjoyed it, and found it useful. Almost every article I post here takes a few hours of my time to research and write, and I have never felt the need to ask for anything in return.
But now – for the first, and perhaps only, time – I am asking for a favour. If you are a patent attorney, examiner, or other professional who is experienced in reading and interpreting patent claims, I could really use your help with my PhD research. My project involves applying artificial intelligence to analyse patent claim scope systematically, with the goal of better understanding how different legal and regulatory choices influence the boundaries of patent protection. But I need data to train my models, and that is where you can potentially assist me. If every qualified person who reads this request could spare just a couple of hours over the next few weeks, I could gather all the data I need.
The task itself is straightforward and web-based – I am asking participants to compare pairs of patent claims and evaluate their relative scope, using an online application that I have designed and implemented over the past few months. No special knowledge is required beyond the ability to read and understand patent claims in technical fields with which you are familiar. You might even find it to be fun!
There is more information on the project website, at claimscopeproject.net. In particular, you can read:
- a detailed description of the study, its goals and benefits; and
- instructions for the use of the online claim comparison application.
Thank you for considering this request!
Mark Summerfield
0 comments:
Post a Comment