AI’s Hidden Ownership Question: What a New “Worldview” Benchmark and the USPTO’s Revised Inventorship Guidance Mean for Innovators

A story that began in the faith-technology corner of the market has landed squarely in the middle of the questions our clients ask us every week: When a machine creates, who owns the result — and who owns what went into it?

The catalyst was a benchmark released by Gloo, a Boulder-based technology company serving churches, ministries, and religious universities. Its “Flourishing AI” benchmark tested leading large language models on questions of meaning, character, and faith, and reported that today’s frontier models perform poorly on those dimensions — worst of all on questions of faith. In a widely read IPWatchdog op-ed, James Edwards used that finding to make a broader point: every AI model carries a worldview that users never see and never agreed to, and that worldview is assembled from human work — books, scholarship, writing, code, and argument — much of it used without credit or compensation.

You do not have to adopt the op-ed’s philosophical framing to see why it matters to intellectual property lawyers. Strip away the subject matter, and the same throughline runs through nearly every hard AI question in front of the courts and the U.S. Patent and Trademark Office right now: ownership — and its two close cousins, attribution and consent.

A Model’s “Worldview” Is Built From Someone’s Work

An AI model does not develop its outputs from nowhere. It derives them from the corpus it was trained on. That simple fact is the engine behind the wave of copyright litigation testing whether training on protected works is infringement or fair use. Those cases are often described as fights about money, but at bottom they are fights about whether creators must be asked, credited, and paid when their work is absorbed into a commercial system and re-expressed in its outputs.

For now, the law is genuinely unsettled. Courts are working through these questions case by case, and early results have been mixed — turning on the specific works, the specific uses, and the specific records in front of each judge. The practical takeaway for clients on both sides is not to wait for a single grand answer. It is to understand where your works, or your training data, sit on the spectrum from clearly licensed to clearly contested.

The USPTO Draws a Line: Only Humans Invent

While the copyright side remains in flux, the patent side just produced a clear, concrete development worth every innovator’s attention.

On November 28, 2025, the USPTO issued Revised Inventorship Guidance for AI-Assisted Inventions (Docket No. PTO-P-2025-0014), rescinding in full the Office’s February 2024 guidance issued under former Director Kathi Vidal. The revised guidance does three things that matter:

•      It reaffirms that only a natural person can be an inventor. An application naming an AI system as an inventor or joint inventor is subject to rejection. This tracks the Federal Circuit’s decision in Thaler v. Vidal.

•      It restores the traditional “conception” test as the single standard. Inventorship turns on whether a natural person formed a definite and permanent idea of the complete and operative invention — the same test applied to any other invention, with no special AI carve-out.

•      It withdraws the 2024 “Pannu factors” overlay. Those joint-inventorship factors apply only when weighing the contributions of multiple human co-inventors. Because an AI system cannot be a joint inventor, the Office concluded the Pannu analysis simply does not belong in the AI-tool context.

The guidance treats AI as an instrument — analogous to laboratory equipment, software, or a research database — that a human inventor uses. It also implements Executive Order 14179 (“Removing Barriers to American Leadership in Artificial Intelligence”).

Notably, this is not an anti-AI posture. Current USPTO Director John Squires has been vocal that Section 101 should not be wielded as a blunt instrument to exclude whole fields of technology, and his early decisions have leaned toward eligibility for AI-enabled inventions. The synthesis the Office is reaching for is pro-AI and pro-inventor: use the machine freely, but keep the human at the center of the legal claim to what it helps produce.

Why This Matters — and the Open Questions

The through-line connecting the copyright fights and the patent guidance is the same instinct: the law keeps re-anchoring ownership in the human act of creation. On the patent side, the Office chose the traditional conception standard precisely because the alternative — extending inventorship to a system that recombines what people made — would sever the link between creation and ownership that the system depends on.

But the guidance is examination policy, not binding law, and it leaves the hard edge cases open. How much of a claimed concept can an AI tool generate before a human’s “conception” is genuinely in doubt? Does selecting, refining, or approving an AI-generated output — particularly for design patents, where AI can produce ornamental designs from minimal input — satisfy conception? Courts and the Office will be answering those questions for years.

Practical Steps for Companies and Innovators

Whether you are building AI, using it in R&D, or protecting a catalog of creative works, a few concrete moves will put you in a stronger position:

1.      Document human conception. Keep contemporaneous records showing which people conceived each claimed feature, especially where AI tools are in the workflow. If a dispute arises, the record of human conception is your best evidence.

2.      Refresh your inventorship playbook. Update invention-disclosure forms, prosecution checklists, and training materials that still cite the rescinded 2024 guidance or its Pannu framing.

3.      Watch your priority claims. A foreign application naming an AI system as the sole inventor will not support a U.S. priority claim. Make sure at least one common natural-person inventor spans your related filings.

4.      Get your data house in order. Track provenance for training data, and treat licensing, attribution, and consent as risk management. In sectors where trust is the value proposition, verifiable permission can be a competitive asset, not just a compliance cost.

5.      If you are a rights holder, take inventory. Know which of your works have value as training material, decide your licensing posture deliberately, and monitor the copyright litigation that will shape your leverage.

The Oldest Question, Asked of a New Machine

The question of who owns what a person creates is not new — it is written into the Constitution’s authorization for Congress to secure to authors and inventors the rights to their writings and discoveries. What is new is the machine now standing between the creator and the creation. The USPTO’s revised guidance answers one version of that question cleanly for patents: the machine is a tool, and only people invent. The copyright side is still being written.

Clients who build the discipline now — documenting human conception, cleaning up data provenance, and making deliberate choices about attribution and consent — will be the ones best positioned however the remaining questions are resolved.

Jennifer Gordon is a partner at Wood Phillips, where she counsels clients on intellectual property and technology. For guidance on AI-related inventorship, patent strategy, or data-licensing questions, contact the author or your Wood Phillips attorney.

This post is provided for general informational purposes only, does not constitute legal advice, and does not create an attorney-client relationship. It reflects developments as of July 2026; the USPTO guidance discussed is examination policy and may be superseded by regulation, legislation, or court decisions.

Sources: James Edwards, “Who Owns the Worldview Inside an AI Model?”, IPWatchdog (July 8, 2026); USPTO Revised Inventorship Guidance for AI-Assisted Inventions, Docket No. PTO-P-2025-0014 (published in the Federal Register Nov. 28, 2025).