What It Means
- The Supreme Court’s judicial AI governance framework now functions as a pre clearance gate, not a guidance document.
- Only one AI tool, the transcription system Scriptix, currently has approval for use inside Philippine courts.
- Every other AI application needs case by case Supreme Court En Banc approval before it can touch a courtroom.
- Vendors who built compliance and human review into their product from the start now hold a structural advantage over vendors selling raw capability.
- The exposure sits with AI legal tech startups and solo practitioners leaning on consumer AI tools for filings, both now standing outside a widening approval model.
The Philippine judiciary has stopped talking about AI in the abstract. On July 1, Senior Associate Justice Marvic Leonen briefed trial court judges, clerks of court, legal researchers, and stenographers in Baguio on how judicial AI governance actually works in practice, and the answer is narrower than most people assumed. Only one tool, the voice to text system Scriptix, is currently cleared to operate inside a Philippine courtroom. Every other AI application, no matter how capable, needs Supreme Court En Banc approval before it gets anywhere near a judge’s desk.
That is not a caution statement. It is a gate.
Judicial AI Governance Was Built to Filter, Not Just Guide
The Supreme Court En Banc approved the Governance Framework on the Use of Human Centered Augmented Intelligence in the Judiciary on February 18, 2026, under a resolution numbered A.M. No. 25-11-28-SC. That resolution is the origin document for judicial AI governance in the country, and Leonen chaired the working group that drafted it, with Associate Justices Ramon Paul Hernando and Rodil Zalameda as vice chairs. The policy’s ethical core rests on fairness, accountability, and transparency, and Leonen’s later briefings describe six operating principles built on that core: human primacy, explicability, auditability, equal access, institutional sovereignty, and continuous learning.
The drafting process did not happen in isolation. The working group pulled from the Council of ASEAN Chief Justices’ own AI governance guidelines and UNESCO’s guidelines for AI use in courts and tribunals. That means a vendor clearing judicial AI governance in Manila is clearing a bar built to be legible against a regional and international standard, not a purely domestic one. A tool built only to satisfy a local checklist has a harder path than one already built to satisfy ASEAN and UNESCO benchmarks.
Read as policy language, these sound aspirational. Read as procurement criteria, they are a checklist. A vendor pitching an AI tool to the judiciary now has to demonstrate that a human stays in control of the output, that the tool’s reasoning can be explained, that its use can be audited after the fact, and that it does not concentrate advantage with litigants who can afford better technology. That is a compliance burden most AI products were never built to carry, because most AI products were built for speed and scale, not institutional accountability.
Judicial AI governance, in practice, functions as a market filter before it functions as anything else.
Scriptix Sets the Template for What Clears
Scriptix, the judiciary’s voice to text transcription tool, is the only AI application currently authorized under judicial AI governance for court use. Leonen offered it as the working example of what “human primacy” looks like in practice: the tool produces a transcript, but a court stenographer still reviews it for accuracy before it becomes part of the record. The machine assists. The human signs off.
That is the bar every future applicant has to clear, not just technically but procedurally. A tool that cannot show a documented human review step built into its workflow does not get to argue its way in on accuracy alone.
A Standing Committee Now Decides Who Gets In
The resolution does not leave future approvals to ad hoc review. It directs the SC to stand up a permanent committee to guide the development and ethical use of AI in the judiciary, drawing representatives from the legal, technical, and academic sectors. That committee is where the real gatekeeping happens going forward. Every vendor pitch and pilot proposal now funnels through this body before it reaches a courtroom.
A standing committee is a different animal from a one time resolution. Committees make judgment calls, build relationships with vendors who show up consistently, and develop institutional preferences over time. Whoever gets a seat, or builds a working relationship with the people who do, has more influence over what judicial AI governance approves next than the resolution’s text alone would suggest. The written principles are static. The committee’s discretion is not.
The Adjudication Line Leonen Will Not Move
Leonen was direct about where the boundary of judicial AI governance sits. AI may assist in research, scheduling, drafting, and analytics, but it may never adjudicate. His reasoning goes past institutional caution and into how these systems actually reason. AI models rely on induction, predicting likely outcomes from historical patterns, and Leonen argues that method fails the law on two fronts: no volume of past cases can guarantee the next one, and the historical record those systems learn from was never a neutral account of anything.
He extended the point further, noting that much of the legal system, rehabilitation, probation, restorative justice, is built on the premise that a person is not identical to their past conduct. A prediction engine trained on precedent cannot hold that premise. It can only repeat it.
A Second Institution Builds the Same Kind of Gate
The judiciary is not the first Philippine institution to formalize AI governance this year. The National Privacy Commission moved earlier in 2026 to tighten its own rules around data scraping and AI compliance, coordinating with regional privacy regulators on enforcement. Those rules work after the fact, flagging violations once they occur.
Judicial AI governance works before the fact. Nothing gets deployed without clearing the gate first. Two different regulators, two different mechanisms, but the same direction of travel: Philippine institutions are moving away from reactive AI oversight and toward pre approval as the default posture. Any enterprise selling AI into Philippine government now has to plan for review cycles, not just sales cycles.
The Bar Judicial AI Governance Sets, and Who Clears It
The vendors positioned to benefit are the ones already selling compliance as a feature: documented audit trails, human in the loop workflows, explainable outputs. That has been a hard sell in a market that mostly rewards speed. Under judicial AI governance, it becomes the entry ticket.
The vendors and users left outside are less visible but more numerous. General purpose AI legal tech startups that assumed government would adopt commodity tools the way private firms did now face, under judicial AI governance, a review process built for institutions, not sales pipelines. Solo practitioners and small firms leaning on consumer AI tools to compensate for thin research staff sit in a similar bind, operating in a gray zone judicial AI governance will increasingly be read to restrict.
The vendors already building audit trails and human review into their products are entering a market with less competition than the one they trained for, because judicial AI governance rewards vendors who planned for institutions, not sales cycles. The ones built for speed have paperwork to do first.
Track more regulatory shifts that affect your business in Policy & Regulation section of Hemos PH.
