AI Patent Applications Have a New Secret Weapon

Authored by Babak Akhlaghi on September 19, 2025.   The USPTO’s August 2025 memorandum represents a significant shift in how examiners evaluate AI and software patents.

Most patent practitioners are focusing on the wrong aspects of this guidance update. While it clarifies existing rules rather than creating new ones, the practical implications for AI patent applications are substantial.

I’ve been writing about Section 101 eligibility law for over 10 years, and I still struggle with its complexity. Judges struggle with it too. One case gets held patent ineligible while another with similar facts gets approved by different judges.

This memorandum provides much-needed clarity in two critical areas that could meaningfully improve outcomes for AI patent applications.

The Hidden Problem Most People Don’t See

Here’s what’s been happening behind the scenes. Examiners have been rejecting AI patent applications by claiming they recite abstract ideas because they “can be performed in a human mind” or they recite “mathematical formula.”

I’ve seen examiners argue that training machine learning models using updated training data is simply a mathematical formula. Even when the examiner admits the claim doesn’t describe a mathematical formula, they’ll assert it’s “inferred” and therefore abstract.

The rejection becomes so problematic that you have to appeal it. This costs clients significant money for expensive appeals on issues that shouldn’t exist.

The Two Game-Changing Clarifications

The memorandum makes two critical clarifications that practitioners are underestimating.

First, it reminds examiners that Step 2A, Prong 1 has real teeth. Claims that merely involve an abstract idea are not patent ineligible. The claim must actually set forth or describe the abstract idea to fail.

The difference is crucial. Example 39 shows “training the neural network in a first stage using the first training set” doesn’t recite a judicial exception because it doesn’t describe specific mathematical relationships. Example 47’s “training the ANN using a backpropagation algorithm and a gradient descent algorithm” explicitly references mathematical calculations, thereby reciting an abstract idea.

Second, patent eligibility rejections must be supported by preponderance of evidence. If it’s a close call, examiners can only make rejections when it’s more likely than not that the claim is ineligible.

A rejection should not be made simply because an examiner is uncertain about eligibility.

Your New Weapon Against Bad Rejections

Patent practitioners now have a concrete weapon to challenge examiner uncertainty.

In addition to using existing tools in my toolbox, I plan to directly challenge examiners’ burden of proof like this: “The Office has failed to meet the preponderance of evidence standard required for a 101 rejection as clarified in the August 2025 memorandum. The examiner’s analysis relies on speculation and uncertainty rather than demonstrating that it is more likely than not that the claim is ineligible.”

When examiners use language like “may involve” or “could be considered,” they’re demonstrating the very uncertainty that should preclude rejection under 101.

Where there is doubt about eligibility, that doubt must be resolved in favor of the applicant.

The Practical Framework That Works

The memorandum clarifies when AI processes can be considered “practically performed in a human mind,” giving patent drafters a concrete test to work with.

My litmus test is simple: Could a human actually sit down with pen and paper and perform this limitation in any reasonable timeframe with any reasonable accuracy?

For “analyzing patterns in data,” focus on volume and complexity. Analyzing patterns across millions of data points in real-time clearly can’t be practically performed in a human mind. But “analyzing patterns” without more specificity gives examiners room to argue humans could do basic pattern recognition.

For “generating predictions,” emphasize the mechanism and data requirements. Generating predictions based on processing vast datasets through trained neural network weights and parameters is practically impossible for humans.

The key is being specific enough in claim language to make that impossibility obvious to the examiner.

Why This Represents a Genuine Policy Change

This memorandum is part of coordinated pro-patent policy changes across USPTO divisions.

The USPTO recently increased discretionary denials in IPRs, which have been used effectively to invalidate issued patents. They’re also strictly enforcing Rule 42.104(b)(4), meaning IPR petitioners must now identify exactly where each claim element appears in prior art. No more expert testimony to fill gaps. No more common sense arguments. No more hand-waving.

Historical data supports optimism about policy clarifications. The 2019 USPTO guidance changes resulted in a 25% decrease in Alice-affected technology rejections and a 44% reduction in uncertainty about eligibility determinations.

This August memorandum appears positioned for even greater impact.

The Window Is Open Right Now

I suspect we’ll see substantially fewer patent eligibility rejections for AI-related applications, which means allowance rates will substantially increase. Section 101 rejections are among the toughest to overcome.

My advice to innovators: seize this moment. If you’re on the sideline about filing an application, file it now.

The biggest mindset change needed is moving from defensive to offensive thinking. Instead of asking “Can this be patented?” and assuming the answer is probably no for AI inventions, flip that question to “How can I describe this invention so it’s clearly patentable?”

Focus on details about neural network architectures, training processes, and data processing volumes. Stop self-censoring promising AI innovations because you think they won’t be patentable.

The innovators who act quickly and strategically in this new environment will build significant patent portfolios while others are still operating under old assumptions.

The window is open. The question is whether you’ll walk through it.

Stay Informed with Expert Insights!

Want more valuable tips on navigating patent law and protecting your intellectual property? Join our free weekly newsletter for the latest updates, expert advice, and exclusive content straight to your inbox. Subscribe today and never miss out on essential information that can make all the difference in your patent journey!  Subscribe to our Newsletter