AI patent review

What can we learn from OpenAI’s patent portfolio?

I get asked about 10 times a week how to protect AI-based technologies and products. It’s a tricky question, because patenting software inventions can be very tough in the US (and Europe), as the patent office dislikes providing protection for abstract ideas.

But let’s simplify the question by just looking at what the much-discussed OpenAI has done in this space. To date, OpenAI has 12 granted US patents. We don’t know how many more are pending, likely because those applications are still unpublished. What we can say is that they don’t seem to be pursuing protection outside of the US for their technology.

So what have they got protected and what can we learn from it?

From a technical review of the portfolio, the patents have several key themes that likely enabled them to get granted. The primary focus across these patents is on technical improvements to computer functionality rather than just abstract ideas. For example, the ‘adaptive UI’ patent (US12164548BA) focuses on providing a dynamic user interface based on the output from the AI. The ‘multimodal interaction’ patent (US12039431BA) introduces “contextual prompts” on images (e.g., clicking on a specific area or annotating it). The ‘LLM interaction’ patent (US12051205BA) moves the cursor to perform actions on the GUI based on the user’s textual prompt (likely relevant to OpenAI’s ‘Operator’ product). The rest of the granted patents relate to the image generation and speech recognition technologies now familiar in the OpenAI product lineup.

The how and what

The common thread across these patents is that they focus on the how rather than the what when describing specific technical implementations and concrete improvements to computer functionality.

To use an analogy, they aren’t trying to protect “a taller tower”. They are protecting “making a tower stronger by putting special pieces at the bottom, and connecting them in this new way”. They are explaining the technical details of exactly how they made the AI product work better. Because they focused on the how, they were able to get their patents granted, as they were not chasing abstract ideas – they were protecting the building blocks that make their technology novel.

If you want to apply the OpenAI patent strategy, you need to do the same. Don’t just say WHAT your AI product does, explain exactly HOW you made the product work better in a new and special way.

Do this, and you too could have a valuation of $340 billion (maybe).