Patent offices worldwide are drowning. The U.S. Patent and Trademark Office reports backlogs exceeding 600,000 pending applications. Policy makers and firms are scrambling to discuss examination timelines, AI-assisted review tools, and fee structures. On the surface, it looks like a capacity crisis.
Look deeper, and you find something more unsettling: we're experiencing a fundamental philosophical breakdown about what intellectual property actually protects.
The tactical conversation is straightforward enough. Industry groups want faster patent grants. The USPTO wants resources. Some argue that AI can help examiners work through applications more efficiently. Others warn that cutting corners invites weak patents that clog courts for years. These are legitimate operational concerns.
But beneath this noise sits a harder question that nobody quite wants to say aloud: If machines can now generate inventive outputs, write code, design compounds, and conceptualize solutions with minimal human direction, what exactly are we granting exclusive rights over?
The patent system was built on a specific bargain. An inventor discloses something truly novel in exchange for temporary monopoly rights. The public gains access to that knowledge eventually; the inventor gains reward and incentive. This assumes humans are the creative agents. It assumes intentionality. It assumes we can identify the person or entity deserving protection.
Generative AI scrambles all three assumptions.
Consider a pharmaceutical researcher who uses an AI tool to propose a novel molecular structure. Is that an invention? Who invented it? The researcher for knowing what problem to solve? The engineer who built the model? The data scientists whose training sets shaped its outputs? The company that owns the software? The answer isn't obvious, and it varies by jurisdiction in ways that make international IP strategy nearly impossible.
Some firms are already filing patents claiming AI-generated inventions with humans listed as "inventors" in name only. Others are experimenting with putting the AI itself as an inventor, which most patent systems explicitly forbid. A few jurisdictions are quietly considering workarounds. Nobody wants to say it clearly because the implications are too large.
If we grant patents on AI-generated work without resolving ownership questions, we've essentially created a legal fiction. The patent appears to protect innovation, but it actually protects whoever controls the computational process. That's not the same as protecting inventors. It's protecting infrastructure.
Meanwhile, the patent backlog persists partly because examiners must now determine novelty in fields where the definition of "novel" is slipping. Is an output novel if it's statistically unlikely but mathematically derivable? If an AI sees something humans never recorded but our data suggests we should have seen, is that invention or inference? These aren't questions the Patent Act anticipated.
The policy response so far has been incremental. Faster processing. Better tools. Clearer guidance on AI-related applications. Sensible, measured steps.
But they miss the structural issue. We're trying to manage a capacity problem when we actually face a legitimacy problem. We need consensus on what the patent system is meant to reward in an era where "creation" increasingly means orchestrating computational processes rather than originating novel concepts.
That conversation is harder than discussing fee schedules. It requires intellectual honesty about who benefits from the current system and who might lose. It requires rethinking whether patents should even apply to certain kinds of AI-generated outputs.
Until that conversation happens, faster backlogs and more examiners just mean we'll grant more patents on inventions we don't yet understand, to entities we haven't yet identified, for rights we can't quite defend in court.
The real story isn't capacity. It's clarity.