The AI arms race is already underway. Most of the attention is on which company, profession, or country will gain the advantage. But some of the most consequential contests may not be one side against another at all. They may be one generation against the next.

We do not know how large the step between generations will be. But if capability continues to compound, a contest separated by several years may not involve comparable tools. That possibility deserves attention wherever one party must act long before the other.

The imbalance is most severe when the first party must commit early with incomplete information, its most important decisions become difficult to revise, and the real test is delayed. By the time the second party acts, it has the finished result, the accumulated evidence, and a newer generation of intelligence. It may also have a much easier job: instead of solving the entire problem, it may need only to find one consequential weakness.

Patents fit every one of those conditions.

There is enormous enthusiasm around AI-assisted patent drafting. Faster applications, lower cost, greater output, and better access to technical knowledge all make the technology look like an obvious win. But patents are not judged at the moment they are written. They are judged later, under pressure, by people who know exactly what they are trying to defeat.

The intelligence available at that moment may be far beyond the intelligence used to create the asset. That should change how we think about “good enough.” Good enough for today’s review process may not be good enough for tomorrow’s adversary. Good enough to issue may not be good enough to survive. And good enough to look sophisticated may not be enough to withstand a system that can search more deeply, generate more alternatives, reconstruct more technical pathways, and test more legal theories than any current team could economically pursue.

The uncomfortable truth is that we do not know where the future bar will be. But it is hard to believe it will be lower. The rush to make patent drafting faster and cheaper may be happening at the same moment the eventual standard of scrutiny is rising.

That is the caution flag. The question is no longer merely whether AI can help produce patents more efficiently. It is whether the patents produced today will be strong enough to survive a later generation of intelligence. Because that contest may arrive years after the critical drafting decisions are locked in, there may be little room for correction once the answer becomes clear.

The same temporal asymmetry applies to business models. A company may optimize around today’s costs, workflows, and technical limits, then face a competitor a few years later whose economics are built on a newer generation of AI. That dynamic is not new. The velocity is.

When capabilities change this quickly, the distribution of possible outcomes widens. Strategies that look durable under today’s assumptions may become far more fragile under tomorrow’s. Patent investment should be viewed through the same lens.

If future systems make attack cheaper, deeper, and more persistent, then the case for patent protection may need to be reassessed. Not abandoned, but re-underwritten.

The prudent response is to treat AI not only as a way to reduce the cost of creating patents, but as a reason to raise the standard for what deserves to be created, how it is reviewed, and how confidently its future value is estimated.

**So while the exact future is uncertain, the asymmetry is not.