The Rule Everyone Quotes That No Regulator Actually Wrote
The Intellectual Property Office of the Philippines (IPOPHL), the government agency that administers patents, trademarks and copyright registration in the country, held public consultations on September 10 and 11, 2026 on a draft framework that sorts AI-touched creative work into four tiers of protectability rather than a single yes-or-no test, according to IPOPHL’s own announcement and reporting by BusinessMirror. That framework is more granular than the rule most licensing teams currently rely on — the assumption that anything an AI system had a hand in simply cannot be copyrighted.
What Manila Actually Drafted: Four Tiers, Not One Gate
IPOPHL’s draft Guidelines on the Registrability of AI Systems and AI-Related Works classifies output into four categories, per the agency’s press release: AI-assisted works, where a human author uses AI as a tool but supplies the original expression; AI-enhanced works, which begin as human-created material later modified with AI; hybrid works, which combine human-authored and AI-generated elements with protection limited to the qualifying portions; and purely AI-generated output, which receives no protection when no human creative contribution meets the originality standard.
Director General Teodoro C. Pascua told stakeholders the guidelines are not meant to regulate AI generally or settle the broader copyrightability debate — only to give applicants clarity on how IPOPHL will evaluate a registration request. Consultations were still open as of September 17, 2026, with no closing date announced, according to trade reporting on the draft.
Where the Binary Rule Came From — and Why It Undersells the Problem
The “AI content can’t be copyrighted” shorthand traces to two U.S. Copyright Office decisions: the 2023 refusal to register “A Recent Entrance to Paradise,” an image generated solely by Stephen Thaler’s AI system without human authorship, upheld by the D.C. Circuit; and the Zarya of the Dawn ruling the same year, which stripped protection from Midjourney-generated images inside a graphic novel while preserving copyright in the author’s text and in her “selection, coordination and arrangement” of the images. Both cases test the same narrow question — did a human do anything beyond typing a prompt — and both said no.
What that binary framing skips is the case IPOPHL’s four tiers are built to catch: work where AI generated raw material but a human then selected, edited, reordered or substantially reworked it. The U.S. rule already allows protection for that “selection and arrangement” layer — Zarya of the Dawn proves it — but licensing teams routinely repeat the simplified version of the rule as if editing an AI draft were legally identical to prompting one. IPOPHL’s draft turns that buried nuance into an explicit, named category instead of leaving it to be rediscovered case by case.
What a Brand or Tech-Transfer Team Should Document Starting Now
Under a four-tier test, the evidence that determines which tier a work falls into is created — or lost — at the moment the work is made, not when a registration dispute arises later. A marketing team running AI-assisted campaign copy through several rounds of human rewriting has a materially different claim than one publishing first-draft AI output unedited, but only if it can show its work. That means version histories, prompt logs, and a record of what a human changed and why are no longer just good practice; under a tiered system, they are the evidence a registrar or a court would need to see.
The stakes are rising with volume. IPOPHL logged a record 53,231 patent, trademark, utility model and industrial design filings in 2025, up 2 percent from 52,257 in 2024, and Pascua told reporters on September 19, 2026 that the office expects to beat that record this year, according to BusinessMirror. A growing share of that pipeline will involve AI at some stage of creation — meaning the categorization question this draft answers will apply to more applications, not fewer, by the time any final version takes effect.
For a tech-transfer office or brand team operating across ASEAN, the immediate move is not to wait for IPOPHL’s guidelines to be finalized. It is to start keeping the documentation a four-tier test would require, regardless of which jurisdiction eventually reviews the filing — because the alternative to a hybrid or AI-enhanced classification is the tier with no protection at all.
Key takeaways
- IPOPHL’s draft guidelines, consulted on September 10–11, 2026, replace a binary “AI content is or isn’t copyrightable” test with four categories: AI-assisted, AI-enhanced, hybrid, and purely AI-generated.
- The binary rule most teams cite comes from two 2023 U.S. Copyright Office decisions that tested only whether a human did more than submit a prompt — a narrower question than IPOPHL’s tiers address.
- Protection under the draft framework depends on evidence created at the time of authorship: prompt logs, edit histories, and records of human selection or arrangement.
- IPOPHL logged a record 53,231 IP filings in 2025, up 2 percent from 52,257 in 2024, and expects to exceed that record in 2026, meaning AI-touched filings will keep growing in volume before the guidelines are finalized.
- Brand, licensing, and tech-transfer teams should start documenting human contribution to AI-assisted work now, rather than waiting for a final rule to tell them what evidence they should have kept.