In my recent livestream, we walked through two exciting new patents from OpenAI. Here’s a quick recap of what makes these patents worth your attention.
Custom GPTs — Building Tailored AI Models
The first patent is titled “Systems and Methods for Generating Customized AI Models” (US 12,406,207 B2). It’s the patent behind OpenAI’s Custom GPTs feature.
At its core, the invention claims a system that:
- receives a query to generate a custom AI agent,
- configures it with tailored knowledge, instructions, and capabilities, and
- deploys it to answer user queries more efficiently than a general model.
What’s exciting here is how relatively broad the claims are. The patent doesn’t just cover one implementation — it aims to capture the entire idea of letting end users “spin up” their own specialized AI models. That breadth raises interesting questions around enforceability: could competitors build similar “custom agents” without running into this patent?
For a detailed discussion of the scope of the granted claims, check the livestream recording.
Another noteworthy detail: this patent is a continuation from a 2024 filing and was granted under Track One accelerated examination. That shows OpenAI’s deliberate strategy of building a fast-moving, protective patent portfolio in parallel with product rollouts.
Multi-Agent Shared Workspaces
The second patent is titled “Multi-Agent Interactions Using a Shared Workspace” (US 12,405,822 B1). The core idea is a shared digital workspace that acts as a ledger of commands. Multiple AI agents can:
- join the workspace,
- view its current state,
- post new commands, or
- decide to “yield” (i.e., explicitly not act when another agent is better suited).
That last point — yielding — is the core of this invention. It introduces a coordination mechanism where specialized agents don’t step on each other’s toes. Instead of all agents racing to act at once, they can recognize when it’s better to hold back.
The claims capture this mechanism relatively broadly, covering the general workflow of agents acting or yielding within a shared environment. For companies building multi-agent systems, this is a patent to study closely.
For more details on the exact claim scope, watch the video recording or the audio podcast.
Putting an Invention into ChatGPT?
During the livestream, Olli Pekonen raised a great question:
My short answer: if you use a public, cloud-based GPT, you should make yourself very familiar what happens to your inputs and make an informed decision. If you’d like to go deeper into how to safely and effectively use AI in your patent work, join me in my next seminar.
Next: Anthropic’s Foundational Patent
I had also planned to walk you through a brand-new patent from Anthropic — their very first US patent — but we simply ran out of time in the livestream. It’s an unusually long document (over 340 pages!), which signals that Anthropic is trying to lock down a broad swath of ground around multimodal agent orchestration.
I’ll definitely return to it in a future session. If you don’t want to miss that deep dive — and other fresh AI patents as they come out — make sure to follow me on LinkedIn so you’ll get notified when the next livestream goes live.