In 1996, Judge Frank Easterbrook famously argued that emerging technologies, like the internet, do not need bespoke legal rules. His essay, Cyberspace and the Law of the Horse, suggested that general legal principles are sufficient, just as there is no "law of the horse" despite horses' historical significance.

Three years later, Professor Lawrence Lessig countered, asserting that cyberspace might reveal new regulatory lessons, such as how the underlying architecture of software code can act as a form of regulation by constraining or directing user behavior. For example, much like physical barriers in the real world constrain actions, algorithms on social media platforms can nudge users toward certain content, potentially justifying tailored legal approaches where existing laws fall short.

Today, the "Law of the Horse" debate is resurfacing in AI and patent law: is a specialized legal framework needed for AI patents, or can existing patent law evolve to meet new challenges?

The Broader AI Law Debate

Policymakers and scholars are divided. The European Union's AI Act, for example, takes a risk-based, AI-specific regulatory approach, imposing obligations tailored to the potential harm of different AI systems, while building on and referencing general EU laws such as data protection under the GDPR, product safety frameworks, and consumer rights directives without prejudice to those regimes. By contrast, the United States has generally adapted existing laws, relying on agencies such as the FTC and OMB to provide guidance, thus far avoiding the creation of an entirely new "AI law".

This divide mirrors Easterbrook vs. Lessig: should AI be governed by a distinct framework, or does adaptation of existing law suffice? Some scholars question whether "AI law" even exists as a distinct field, viewing it instead as applied technology law. Others argue against new "AI law" acronyms, asserting that existing regimes in privacy, civil rights, and consumer protection are adequate, with accountability placed on AI deployers rather than creating overly prescriptive rules.

Patents in Focus: Challenges with AI

AI patents bring practical challenges that test the limits of existing IP law:

  • Inventorship: Recent decisions, such as Thaler v. Vidal, 43 F.4th 1207, 1210 (Fed. Cir. 2022), underscore the complexity of attributing inventorship when AI contributes to the inventive process, with consistent rulings across the U.S., EU, UK, and Australia affirming that only natural persons can be inventors. This aligns with the USPTO's revised November 2025 guidance on AI-assisted inventions, which treats AI strictly as a non-inventive tool (similar to lab equipment or software), requiring humans to fully conceive the invention under traditional standards.
  • Patent Eligibility: Applying the existing Alice/Mayo framework under §101 to AI claims presents ongoing challenges. The USPTO's July 2024 guidance, updated via 2025 memos on judicial exception reminders and precedential PTAB decisions, requires assessing if claims are directed to ineligible exceptions like abstract ideas (e.g., mathematical concepts, mental processes, organizing activity), then determining eligibility if integrated into practical applications via technological improvements beyond generic use. Variability can arise from over-reliance on mental process categorizations for complex AI, inconsistent weighting of evidence for technical advancements (e.g., improved processing), and uncertainty in AI/ML contexts, where what qualifies as non-routine, tangible innovations can be unclear.
  • Enablement: AI inventions often require detailed disclosures to allow replication without undue experimentation, a challenge noted in both academic and practitioner analyses, especially for black-box models. However, recent PTAB decisions affirm that, even for AI inventions, the level of detail needed depends on the knowledge of a person of ordinary skill in the art, and algorithm-level specifics may not always be required if the specification enables practice without undue experimentation.
  • Obviousness: AI-augmented invention may elevate the level of ordinary skill in the art, requiring patents to demonstrate human creativity beyond AI's routine capabilities. Existing factors from cases like Environmental Designs, 713 F. 2d 693 (Fed. Cir. 1983), and KSR, 550 U.S. 398 (2007), can reasonably be applied to evolving levels of skill using AI in new technical fields without new doctrines.
  • Prophetic/AI-like Disclosures: The pending petition before the Supreme Court in Agilent Technologies, Inc. v. Synthego Corp. (No. 25-570) could reshape prior art presumptions if certiorari is granted, requiring demonstrable functionality rather than hypothetical descriptions in prior art references. Prophetic disclosures refer to hypothetical examples in patent applications, which have existed prior to AI but are likely to proliferate with increasing AI-generated content, suggesting that broadly-applicable doctrines can handle such overlaps without requiring a new framework.

These factors create uncertainty for innovators: while patents remain foundational to IP strategy, AI's unique characteristics strain traditional doctrine.

Pros and Cons of a Custom AI Patent Framework

Weighing the advantages and disadvantages of developing a custom framework for AI patents reveals a tension between providing targeted clarity for emerging technologies and preserving the flexibility of established legal principles to foster broad innovation.

Pros:

  • Greater clarity on inventorship, eligibility, and disclosure requirements
  • Tailored protections for AI-specific innovation, encouraging R&D and investor confidence
  • Alignment with global regulatory trends, particularly in regions like the EU
  • Addressing unique aspects of AI's data-driven and multifaceted nature, potentially requiring a more holistic regulatory view

Cons:

  • Regulatory complexity and higher compliance burdens, which could slow innovation and favor large incumbents
  • Potential fragmentation across jurisdictions, creating challenges for global companies with AI products
  • Risk of over-prescription, limiting flexibility in adapting existing patent strategies
  • A custom framework might become outdated quickly as AI evolves, requiring frequent revisions and potentially discouraging rapid iteration due to lack of clarity

Analyses suggest established guidelines are sufficient in many areas for now, allowing for sufficient certainty without a full overhaul. A hybrid approach—clarifying AI's application under existing patent law while integrating guidance where gaps exist—may strike the right balance, as patent law "must (and can) adapt" to AI-augmented invention. For instance, current law is adaptable in areas like obviousness, where existing case law on PHOSITA can incorporate use of AI as a routine tool. However, AI may raise unique issues for inventorship, such as precisely defining 'significant human contribution' in AI-assisted conception.

Looking Ahead

For AI innovators, 2026 is shaping up to be a pivotal year for patents: courts and the USPTO are issuing clarifying guidance, while ongoing litigation and potential Supreme Court decisions may test doctrines around inventorship, enablement, and prior art. For example, the USPTO has sought comments and may implement new guidance on AI's impact on the level of ordinary skill in the art. Legislative pushes, such as the proposed Patent Eligibility Restoration Act (S. 1546), seek to clarify patent eligibility by replacing judicial exceptions with specific statutory exclusions, potentially expanding eligibility for AI-related inventions.

Strategic planning now—grounded in strong patent fundamentals and awareness of AI's unique contributions—will determine which companies maintain defensible, investor-ready IP portfolios. Practical steps include auditing AI use in R&D, documenting human contributions (e.g., prompts and selections from AI outputs), and considering trade secrets for unpatentable AI elements.

Judge Easterbrook's "Law of the Horse" metaphor reminds us that AI may challenge existing doctrines. Yet, a hybrid approach that leverages the adaptability of current law, while filling gaps where AI creates novel issues, may provide the most resilient framework, emphasizing adaptation over wholesale customization.