If your category is regulated, “can AI make this?” is the wrong question. It can. The question your legal team will actually ask is different: can you prove how it was made, and can you prove it a year from now?
That is a production question, not a technology one. This page is how we answer it.
What "brand-safe" means once the category is regulated
For most advertisers, brand safety is an adjacency problem — don’t let the pre-roll run against something ugly. In a regulated category it means three separate things, and AI only complicates the third.
- What you may claim. Governed by your regulator, unchanged by how the asset was made.
- What you may show. Governed by your code — age, setting, consumption, imagery.
- How it was made, and what you can prove about that. This is the new one. Traditional production answers it with call sheets, releases and invoices. AI production has to answer it deliberately, or it does not answer it at all.
Most AI production problems in regulated categories are failures of the third kind. Nothing in the ad breaks a rule. Nobody can produce the paperwork showing why it doesn’t.
The Philippine layer
Advertising in the Philippines is screened before it runs. The Ad Standards Council — a self-regulatory body formed on 31 March 2008 — reviews and approves advertising material prior to placement or broadcast, across television, radio, print, internet, out-of-home and cinema. Political advertising is the notable exclusion.
For our clients’ categories the layer above that is statutory:
Tobacco. Republic Act 9211 phased in a total mass-media advertising ban. Television, cable and radio from 1 January 2019. Cinema and outdoor from 1 July 2019. All remaining mass media from 1 July 2020, with a single exception for material inside point-of-sale premises. Still in force in 2026. In practice this is why nicotine-category work moves to owned channels, events and export markets — there is no domestic broadcast route left to buy.
Alcohol. FDA Circular No. 2019-006 governs commercial display, selling, promotion and advertising. Promotional material must make clear the beverage contains alcohol and must not be promoted for sale to or consumption by minors; packaging and labelling must not appeal to children. Violations carry a fine of ₱50,000 to ₱100,000.
The gap that matters
Here is the finding. There is currently no Philippine rule on AI-generated advertising.
The ASC’s 2026 circulars run from 2026-004 to 2026-013. They cover holiday operating schedules, a revised music policy, the launch of a virtual presentation feature in the ASC portal, two circulars on HUHS product applications, and one on the advertising materials of gambling operators. Not one addresses AI, synthetic media, or disclosure of generated content.
So there is no local box to tick. The obligation does not disappear — it falls back onto the general truthfulness provisions of the Code of Ethics, onto your brand’s own global AI policy, and onto the rules of every market the asset is exported into. Which is where it gets expensive.
Your Philippine-made asset lands under someone else's rules
Most work we produce out of Makati runs across APAC, North America, Europe, and the Middle East & Africa. The strictest rule in that footprint sets the standard for the master, because re-versioning a finished campaign for one market’s disclosure requirement costs more than building it correctly once.
As of 2 August 2026 — three weeks ago — Article 50 of the EU AI Act applies. Deployers who use AI to generate or manipulate image, audio or video content constituting a deepfake must disclose that the content is artificially generated or manipulated. Disclosure must reach the viewer at first exposure at the latest, in a clear and distinguishable manner, perceivable without special tools — a visible or audible label. Content that is evidently artistic, creative, satirical or fictional gets a lighter obligation: still disclosed, but in a manner that does not spoil the work.
Non-compliance is fineable up to €15 million or 3% of worldwide annual turnover.
Note the word deployers. That obligation attaches to the brand running the ad, not only to the model vendor. If we hand you an asset with no provenance record, we have handed you your compliance problem.
Chain of custody: what should exist before the asset ships
This is the part that separates a production partner from a prompt service. For every regulated deliverable there should be a record you can hand to legal without a scramble:
- Which models produced which elements, at what version, on what date. Model behaviour changes between versions; “we used AI” is not an answer twelve months later.
- What went in. Reference material, brand assets, and their licences. The provenance of the input governs the ownership of the output.
- Which elements are human-authored and which are purely generated — recorded at the time, not reconstructed. This is the distinction copyright turns on, and nobody can rebuild it from memory after the fact.
- Talent consent covering digital replication specifically, where any real person’s likeness or voice appears. A standard shoot release was not written for this.
- The disclosure decision and its reasoning, per market. Including the markets where you decided none was required, and why.
- A named human approver at each gate. Regulators and clients both ask who signed off.
None of this is exotic. It is the traditional production paper trail, rebuilt for a pipeline where the “shoot” leaves no natural residue.
Provenance: use C2PA, and know where it stops
The Coalition for Content Provenance and Authenticity is now the de facto standard — over 6,000 members and affiliates as of January 2026. Google is rolling Content Credentials verification across Gemini, Search and Chrome. OpenAI attaches C2PA metadata to supported generated media.
Attach credentials. Then do not rely on them alone.
The honest limitation is preservation: uploads, screenshots, exports and platform transformations routinely strip or break the metadata. C2PA is a provenance signal, not proof. By the time your asset has been through a social platform’s transcoder it may carry nothing at all.
Which is why the internal record above is the real artifact and the embedded credential is the convenience. Anyone selling you C2PA as a compliance solution has not tested what survives an upload.
Where we tell brands not to use AI
Refusing work is part of the offer. We advise against generated assets when:
- The claim needs substantiation that only a real test can produce. Efficacy, before-and- after, performance demonstrations. If a regulator asks for the underlying evidence, a generated depiction is not evidence — it is the thing being questioned.
- A real, identifiable person is endorsing. Endorsement is a personal act. Synthesising it is the failure mode with the fastest path to a complaint.
- The category’s rules turn on documentary accuracy — anything where the audience is entitled to believe the depiction is a record of something that happened.
- Disclosure would undercut the creative premise. If the idea only works when the viewer believes it is real, and the market requires you to say it is not, the idea is wrong for the channel. Better to find that in the brief than in the clearance queue.
There is nothing brand-safe about a campaign that clears legal and loses the audience’s trust.
How PUNX works in these categories
We produce commercials and brand content for Procter & Gamble, San Miguel, Petron and JTI — the kind of brands whose guidelines don’t bend — with AI-driven workflows out of Makati, for teams across APAC, North America, Europe, and the Middle East & Africa.
That roster is regulated by construction. Nicotine — JTI, on Nordic Spirit Moodsetter and Winston. Alcohol — San Miguel, on GSM Blue. Fuel — Petron, on Petron Sprint. Personal care and household — P&G, across Ariel India, Head & Shoulders, Rejoice Thai, Olay Pan-Asian and Lenor. Quick-service restaurant — KFC.
Between them those categories carry a statutory mass-media ban, an FDA circular, cosmetic claim substantiation rules, and children’s-audience restrictions. The approval workflow is not an add-on we built for this page. It is the only way work in these categories ever ships.
See the work we produce for regulated brands · How our AI video production works · AI vs traditional production
Frequently asked questions
Is AI-generated advertising legal in the Philippines?
Yes. No Philippine law or Ad Standards Council circular currently restricts AI-generated advertising specifically, and the ASC’s 2026 circulars do not address it. Material still requires ASC screening before placement or broadcast, and still has to satisfy the general truthfulness and decency provisions of the Code of Ethics plus any category rules that apply.
Does the Philippines require AI-generated ads to be labelled?
Not at present. There is no local disclosure requirement. If the same asset runs in the European Union, Article 50 of the EU AI Act has applied since 2 August 2026 and requires deepfake content to be disclosed to viewers at first exposure.
Can AI-generated video be used for tobacco or nicotine brands in the Philippines?
The production method is not the constraint — the channel is. Republic Act 9211 has banned tobacco advertising across all mass media since 1 July 2020, with an exception only inside point-of-sale premises. That ban applies regardless of how the material was produced.
Who is liable if an AI-generated ad breaches a regulation — the brand or the agency?
Regulatory obligations generally attach to the advertiser. Under the EU AI Act the transparency duty falls on the deployer — the party putting the content in front of an audience. Your production partner’s job is to give you the record that lets you discharge it.
Does C2PA or Content Credentials make an ad compliant?
No. Content Credentials are a provenance signal and a good practice, not a compliance certificate — and the metadata is frequently stripped by uploads, exports and platform processing. Keep an internal production record independently of what is embedded in the file.
What should a brand ask a production partner for before approving AI work?
The model and version used for each element, the licence status of every input, which elements are human-authored, talent consent covering digital replication, the per-market disclosure decision, and a named approver at each gate.