The Obviousness Question in an AI World

An AI system autonomously uncovers performance issues in a neural network and proposes refinements that materialize from iterative training alone, free from explicit coding. Is the resulting invention obvious?

Such scenarios complicate today's §103 analysis. Obviousness asks whether an invention would have been obvious to a person having ordinary skill in the art (PHOSITA) at the time of invention, evaluated through factual inquiries including determining the scope and content of the prior art, ascertaining the differences between the claimed invention and the prior art, and resolving the level of ordinary skill in the pertinent art. Graham v. John Deere, 383 U.S. 1 (1966). In Environmental Designs, Ltd. v. Union Oil Co., the Federal Circuit emphasized that obviousness must be assessed in light of the capabilities of a skilled artisan—not hindsight reconstruction. 713 F.2d 693, 697 (Fed. Cir. 1983). The Supreme Court later reinforced a flexible approach in KSR v. Teleflex, cautioning that predictable combinations of known elements are generally unpatentable, while rigid rules distort the analysis 550 U.S. 398, 415–22 (2007).

How AI Complicates the Obviousness Framework

AI complicates—but does not upend—the established obviousness framework. As AI tools become embedded in routine R&D, the baseline skill level of PHOSITA in many fields is arguably raised. The risk lies in treating AI-assisted outcomes as automatically obvious simply because AI was involved. Scholars have warned that failing to account for AI's role could lead to systematic over-obviousness. As argued in Ryan Abbott's article Everything Is Obvious, as machine intelligence advances, courts must recalibrate PHOSITA to preserve patent incentives or risk collapsing novelty into inevitability. Similar concerns appear in analyses from scholarly journals that caution against equating AI-enabled invention with routine automation, with one proposing mandatory AI disclosure and heightened obviousness scrutiny for AI-involved inventions to better tailor reviews.

The USPTO's Evolving View of AI and PHOSITA

The USPTO has acknowledged these tensions. In April 2024, the Office issued a "Request for Comments on the Impact of the Proliferation of Artificial Intelligence on Prior Art and PHOSITA," explicitly asking whether AI affects how examiners should assess motivation to combine, reasonable expectation of success, and ordinary skill. Stakeholder responses diverged. Technical organizations emphasized that AI is increasingly a standard research tool, while industry groups warned that redefining PHOSITA as inherently "AI-enhanced" could erode patentability across entire sectors. Pharma stakeholders urged a fact-based, measured approach, noting that AI's impacts remain speculative and cautioning against premature changes that could undermine incentives in high-cost innovation fields.

Courts and Commentators Signal Evolution, Not Revolution

Despite these concerns, commentary largely favors doctrinal evolution rather than overhaul. A legal analysis from IP professionals highlights the flexibility of §103 principles, particularly predictability and evidentiary rigor, but notes a stakeholder divide and the potential need for statutory or PTO guidance revisions to fully address AI's impacts. Another legal commentary similarly argues that Environmental Designs already provides the doctrinal tools to account for AI-augmented skill, provided courts can distinguish routine AI usage from genuinely inventive application.

Academic literature reinforces this view. Analyses in scholarly journals stress the importance of keeping the "person" in PHOSITA, warning that AI should inform—but not replace—the human creativity inquiry. Parallel analyses from scholarly sources highlight that obviousness should turn on how AI is used rather than its mere presence.

To visualize the PHOSITA's potential evolution, consider the following phases as a structured progression:

  • Pre-AI Phase (The Past): PHOSITA is a human with ordinary skill, accessing traditional prior art. This sets the baseline for non-obviousness determinations.
  • AI-Augmented Phase (Now): PHOSITA is a human using AI as a tool (e.g., for massive data analysis or many-variable simulations). This raises the skill level, making more combinations predictable and potentially narrowing patentable space.
  • AI-Dominant Phase (Speculative): PHOSITA could incorporate AI-enhanced or machine standards (as warned by Abbott). This risks an "everything obvious" scenario, though current doctrine limits this by emphasizing human inputs and evidence.

While AI's full impact remains speculative (as echoed in the USPTO stakeholder responses), these phases illustrate the progress of the incremental shifts underway, and where they could be going.

Alternative Perspectives and Proposals

Not all views align with this evolutionary optimism. Critics of Abbott's "everything obvious" prediction argue the risk is overstated—AI still lacks human-like generalization and relies on circumscribed human inputs, preserving a human-centric PHOSITA. Instead, some scholars suggest strengthening utility and prior art standards to protect deliberate human research from speculative AI-generated ideas. Alternative proposals include mandatory AI disclosure, which could enable tailored obviousness reviews but risks discouraging disclosure of AI inventions in favor of trade secret protection. Recent Federal Circuit decisions blur §101 and §103 by treating generic AI as abstract, e.g., Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205 (Fed. Cir. 2025), potentially heightening scrutiny without formal redefinition. But this contrasts with recent USPTO decisions like Ex parte Desjardins, Appeal 2024-000567 (P.T.A.B. Nov. 4, 2025) (precedential), rejecting overly broad application of §101 to categorically reject AI innovations in favor of tools like obviousness "to limit patent protection to its proper scope."

What Recent PTAB Decisions Tell Us

Recent PTAB decisions reflect this incremental approach, continuing to apply traditional obviousness analysis without explicitly redefining PHOSITA to include AI as an autonomous decision-maker. In multiple 2025 trial and appeal decisions involving AI-related claims, the Board has focused on familiar deficiencies: weak prior art mappings, conclusory motivations to combine, and unsupported assertions that AI use was routine.

For instance, in Tesla, Inc. v. Autonomous Devices, LLC, No. IPR2023-01173 (P.T.A.B. Jan. 3, 2025), the PTAB invalidated AI-related claims as obvious based on prior art, emphasizing evidence of predictability. This trend persists into early 2026, as evidenced by a review of 20 recent PTAB opinions on appeals decided from January-February 2026 involving AI/ML-enabled inventions (e.g., involving neural networks, CNNs, GANs, and predictive models) (citations below). These decisions do not define POSITA's skill level explicitly (e.g., as an "AI expert with X years"), instead inferring it from the art and focusing on whether examiners' combination rationales align with what a skilled artisan would reasonably understand or do.

Applicants succeeded (at an even 50% reversal rate) by arguing technical incompatibility, inoperability, undue experimentation, hindsight bias, or domain differences in combining references (e.g., Appeal 2025-002636, prior art combination failed to teach the claimed relationship between classifiers, relying on hindsight to combine disparate teachings; Appeal 2025-002837, prior art failed to teach recommending an alteration of a workflow in response to a comparative analysis, highlighting domain differences in educational modeling versus workflow alteration). PTAB reversed when examiners provided conclusory statements without evidence of how POSITA could integrate elements without hindsight (e.g., Appeal 2025-002256, examiner's mapping was conclusory and the prior art failed to teach gates that are conductively connected together, lacking evidence of motivation for such integration without hindsight; Appeal 2025-003128, examiner's conclusory interpretation of prior art excerpts that did not sufficiently teach determining a model confidence by combining similarity scores).

Affirmations occurred when proposed combinations were feasible (e.g., Appeal 2025-001718, modifying processor with neural network optimization algorithm for more accurate modeling results), enhanced efficiency/accuracy (e.g., Appeal 2025-002784, combining chain-of-custody ledger with fingerprinting for partial representations to improve detection of document alterations), and supported by prior art teachings (e.g., Appeal 2025-002533, integrating prioritized listing with ranking for summaries based on communication session priorities), dismissing inoperability claims if references addressed deficiencies (e.g., Appeal 2025-001523, rejecting argument that modifying predictive model with CNN would render it inoperable, as the combination provides a more robust asset identification technique without changing core principles).

Recent PTAB opinions thus underscore evidentiary scrutiny for rejections based on AI/ML combinations, requiring clear, evidence-based motivation under KSR—aligning with doctrinal evolution rather than overhaul. Absent specific USPTO or court guidance on AI obviousness, PTAB applies traditional PHOSITA analysis rigorously, viewing the skilled artisan as capable of adapting ML tools but demanding explicit rationale for non-obvious integrations. This reinforces warnings of systematic over-obviousness (per Abbott) if rationales remain weak, while echoing calls for evidentiary rigor.

Practice Implications for AI-Augmented Inventions

For practitioners, signals are emerging. While not yet triggering a revolution in obviousness doctrine, AI is raising expectations for how human ingenuity is demonstrated. Drafting strategies and arguments should clearly detail where AI departs from routine tool usage, emphasizing measurable and unexpected improvements where possible to counter obviousness risks. Consider also balancing complete AI disclosure for patentability with trade secret protection for undiscoverable features.

To overcome obviousness challenges for AI-enabled inventions in examination and PTAB proceedings, consider:

  1. Lack of feasibility or interoperability in combining references, supported by evidence of technical incompatibilities or undue experimentation
  2. Conclusory rationale indicating hindsight bias
  3. Domain-specific barriers crossed without explicit teachings
  4. Providing declarations or secondary evidence showing unpredictability of proposed AI integrations

These tactics align with successful reversals in recent cases where examiner rationales lacked specificity.

Litigation strategies provide further opportunities for expert evidence to rigorously test whether alleged motivations to combine reflect true predictability or hindsight gloss. Patent challengers might leverage AI-generated prior art, but this may raise questions of operability unless human-verified, adding another potential layer of evidentiary scrutiny.

Looking Ahead

For now, obviousness in the AI era involves an incremental recalibration rather than a doctrinal reset. But as AI capabilities advance, pressure on §103's elastic concepts will only grow. Monitoring evolving trends in USPTO guidance, PTAB decisions, and judicial precedents will be essential for anyone prosecuting or litigating AI-augmented inventions.

Recent PTAB Appeal Decisions Involving AI-Related Innovations

Obviousness Reversed (in whole or in part):

  • Ex parte Pesic, Appeal 2025-002256 (P.T.A.B. Jan. 22, 2026)
  • Ex parte Naik, Appeal 2025-002636 (P.T.A.B. Jan. 26, 2026)
  • Ex parte Veyseh, Appeal 2025-002286 (P.T.A.B. Jan. 22, 2026)
  • Ex parte Chand, Appeal 2025-002298 (P.T.A.B. Jan. 26, 2026)
  • Ex parte Gao, Appeal 2025-003128 (P.T.A.B. Feb. 4, 2026)
  • Ex parte Karri, Appeal 2025-002837 (P.T.A.B. Jan. 27, 2026)
  • Ex parte Kwatra, Appeal 2025-002969 (P.T.A.B. Jan. 30, 2026)
  • Ex parte Bulut, Appeal 2025-003413 (P.T.A.B. Feb. 5, 2026) (affirmed on §101 grounds)
  • Ex parte Yu, Appeal 2025-003160 (P.T.A.B. Feb. 5, 2026)
  • Ex parte Mihalef, Appeal 2025-003220 (P.T.A.B. Feb. 11, 2026)

Obviousness Affirmed:

  • Ex parte Pradeep, Appeal 2025-001718 (P.T.A.B. Jan. 21, 2026)
  • Ex parte Miller, Appeal 2025-002784 (P.T.A.B. Jan. 29, 2026)
  • Ex parte White, Appeal 2025-002533 (P.T.A.B. Jan. 29, 2026)
  • Ex parte Stahlfeld, Appeal 2025-001523 (P.T.A.B. Jan. 28, 2026)
  • Ex parte Huck, Appeal 2025-003087 (P.T.A.B. Jan. 28, 2026)
  • Ex parte Tran, Appeal 2025-002324 (P.T.A.B. Jan. 30, 2026)
  • Ex parte Egbert, Appeal 2025-002617 (P.T.A.B. Feb. 12, 2026)
  • Ex parte Murali, Appeal 2025-003162 (P.T.A.B. Feb. 6, 2026)
  • Ex parte Bucciarelli-Tieger, Appeal 2025-001008 (P.T.A.B. Feb. 9, 2026) (reh'g denied)
  • Ex parte Ripley, Appeal 2025-002618 (P.T.A.B. Feb. 10, 2026)