On 17 October 2026, at the Konvensi Humas Indonesia 2026 in Surakarta, Deputy Minister of Communication and Digital Affairs Nezar Patria said the ministry (Komdigi) is studying a ministerial regulation under which AI developer companies would label the authenticity of AI-generated video, images and digital media. Infopublik, the government's news portal, reported the remarks on 19 October. MLex described the same proposal as one using metadata-based provenance. This is a proposal. The coverage we could read names no penalty and no issuance date.
Some outlets say Komdigi is "studying" the rule and others say it is "drafting" it. We rely on the wording in the government's own report, which is "studying". The distinction matters because a study can still end in a narrower rule, or in no rule.
The strongest case for labeling
The regulators' concern is real. Patria cited a Hong Kong deepfake video-call fraud that cost a company USD 25 million. Synthetic video is now cheap and convincing enough that ordinary viewers cannot be expected to spot it. A provenance label placed at the point of generation is one of the few interventions that scales. It puts the burden on the few firms that build generators, not on the millions of people who watch the output. A label also does not remove any speech. It adds information.
That is why the idea has moved so widely. The European Union's AI Act, Regulation (EU) 2024/1689, has an Article 50 that requires providers of AI systems that generate synthetic audio, image, video or text to mark outputs in a machine-readable format so they are detectable as artificially generated. Deployers of systems that produce deepfakes must disclose that the content is artificial. Indonesia's proposal, as reported, resembles the provider-side half of that design.
Where the design could go wrong
It is a good sign that the reported proposal targets developers and does not fall on users or on individual posts. The risk is scope creep during drafting. We see four pressure points.
Who is covered is unclear. Reports differ on whether the duty falls on developers only or reaches platforms too. Infopublik mentions only developers. Earlier reporting on Patria's remarks has floated shared detection duties for platforms and developers, but we could not read that article, so we treat it as unconfirmed. A platform duty to detect and label all synthetic content creates a very different problem. Detection is error-prone, and over-labeling real footage as fake, or removing it, hits journalists and citizen witnesses hardest.
Labels are easy to strip. Metadata can be removed by a screenshot or a re-encode. A rule that treats a missing label as proof of authenticity would reward the worst actors, who will not label. Any regulation should say plainly that an unlabeled item is not therefore genuine, and that a label is a signal, not a verdict.
Open-source and small developers. A duty written for large commercial vendors can fall hardest on Indonesian startups and researchers who build on open models. If the rule is to protect local innovation, it should phrase obligations as what is technically feasible and should set clear thresholds so that small developers are not caught by default.
Enforcement. Indonesia's existing electronic-system regime shows how blunt the tools can be. Private operators must register under Ministerial Regulation No. 5/2020, and in September 2026 Komdigi told 25 platforms to register or face enforcement that could end in access blocking, according to MLex reporting on the registration deadline. Attaching that kind of sanction to a labeling duty would turn a transparency measure into a gateway for takedown pressure. Proportionate steps such as warnings, cure periods and fines should come before any blocking.
The compliance load is already heavy
The labeling idea arrives on top of other obligations. Government Regulation 17/2025 on child protection in electronic systems, known as PP Tunas, has applied since 28 March 2026. According to Antara, a Komdigi official said it must apply to all electronic system operators, public and private, and that operators had to submit risk self-assessments by 6 June 2026. Each new rule asks the same firms for another round of documentation, so regulators should check whether labeling can be met through existing registration and assessment channels. Another stand-alone regime is not needed.
What a good rule would contain
First, publish a draft and take comment before issuing. A ministerial regulation can be issued faster than a statute, which makes consultation more important and not less. Second, align with an interoperable provenance standard, as the EU approach does with machine-readable marking, so Indonesian rules do not become a special national format that foreign and local developers must build separately. Third, cover only content a system generates or substantially alters, and exempt ordinary editing. Fourth, state that political satire, parody and journalism are not violations merely because they use synthetic media. Fifth, set a fixed, reasonable compliance period.
The ministry has also said it is preparing a National AI Roadmap for 2026 to 2029 with input from more than 400 bodies, per Infopublik. The labeling rule should be written to fit that roadmap, so that Indonesia gets one coherent AI policy and not a set of overlapping instruments.
Bottom line
A narrow developer-side provenance duty is a reasonable response to deepfake fraud, and it is far better than takedown mandates or criminal penalties for users. The details will decide whether it works. If the final text stays on generators, avoids detection mandates for platforms, and keeps penalties proportionate, it can protect people without chilling speech or local AI building. If it grows into a general content-control power, it will not.