
How do you choose an AI development partner in Indonesia? For most companies the answer depends less on which firm gives the most impressive demonstration than on rules that already decide who answers for an AI system that handles personal data. Indonesia's Personal Data Protection Law, UU 27/2022, keeps the company that hires a supplier responsible for what the supplier does with that data. Its implementing regulation, PP 33/2026, takes effect on 16 January 2027. Komdigi and the Financial Services Authority (OJK) have both published guidance on responsible AI. This guide explains what those documents expect, the main kinds of partner available and the questions worth asking before signing.
What the data protection law expects of a supplier
Under UU 27/2022 the organisation that decides the purpose of processing and controls it is the controller; a supplier that processes data on its behalf is the processor. That is set out in Article 1, points 4 and 5, according to a Hukumonline legal clinic article updated on 23 September 2026. An AI developer that trains, tests or runs models on customer records will usually be a processor.
- The processor must act on the controller's instructions (Article 51(1)), and responsibility stays with the controller (Article 51(3)). Hukumonline notes that, under PP 33/2026, responsibility moves to the processor where it acts beyond those instructions and purposes.
- A processor may bring in another processor only with the controller's prior written consent, according to Hukumonline. This matters when a developer builds on outside cloud platforms or model providers.
- The appointment must rest on a written agreement. Hukumonline says PP 33/2026 sets its minimum contents, including the scope and method of processing, the types of data and audit and inspection.
- SIP Law Firm, writing on 23 October 2026, reports that PP 33/2026 bars clauses that reduce or remove the controller's obligations, so vendor onboarding documents need review.
- SIP Law Firm also reports that new technology such as AI and machine learning triggers a data protection impact assessment, and that transfers of personal data abroad rest on an adequacy finding, safeguards such as standard contractual clauses or binding corporate rules, or consent for limited one-off transfers.
The authority meant to oversee these rules had not been fully formed as of September 2026, SIP Law Firm notes.
What Komdigi and OJK say about responsible AI
Komdigi's AI ethics circular, SE Menkominfo 9/2023, was signed on 19 December 2023 by Budi Arie Setiadi, then Minister of Communication and Informatics. It is addressed to businesses in AI programming, under business classification KBLI 62015, and to public and private electronic system operators. It lists nine values, among them transparency, credibility and accountability, security, personal data protection and intellectual property. It asks developers to provide information about their AI so that harm to users can be prevented, to consider risk and crisis management, and to make sure AI is not made the decision-maker on matters affecting humanity. It is a circular, a form of guidance; the published text lists no sanctions.
For banks, OJK launched its guide to governing AI in Indonesian banking on 29 April 2025. Antara reported that Dian Ediana Rae, OJK's chief executive of banking supervision, said banks should form an AI committee drawing on legal, compliance, risk, data science, cyber security and customer service, either on its own or within the IT steering committee. Infobank, reporting in November 2025, described three core values: reliability, which includes explainability and robustness; accountability, which includes transparency and data privacy; and human oversight. According to Antara, the guide follows AI from initiation and design through model development, testing and implementation to evaluation and periodic audit. One report, in Kontan, gave a February 2025 date for the guide; OJK's own press release gives 29 April.
The kinds of partner, and when each fits
This section is our analysis.
- Large global consultancies and IT services firms bring breadth across strategy, change management and integration, and are used to big procurement processes. They usually suit programmes that span many countries or business units, at a cost and pace that match that scale.
- Cloud and AI platform providers and their certified partners are often the quickest route when the work fits the provider's own tools. The trade-off is closer dependence on one platform, which affects where data sits and how easy it is to leave.
- Local integrators and software engineering firms work closer to Indonesian rules, language and hours, and often suit a defined product that is built and handed over. Depth varies widely, so evidence matters more than branding.
- An in-house team makes sense where AI is central to the business and there is steady work for it. It is slower to start and hard to staff, and many organisations pair it with an outside partner in the early stages.
Questions to ask before you sign
This section is our analysis, not legal advice.
- Where will our data be stored and processed, including by any model or cloud provider you use, and which of those providers are sub-processors?
- Will our data be used to train models that serve other clients, and can the contract exclude that?
- How will you measure quality before launch? Ask to see the test set, the target and what happens when results fall short.
- How are the system's actions and outputs logged, and how can a person review or override a decision?
- What do we own at the end: code, prompts, tuned models, test data and documentation?
- How do we leave? Ask how data is returned or deleted and what it would take to move to another supplier.
- Will you support our data protection impact assessment and, for a bank, the evidence its AI committee will ask for?
Warning signs in a proposal
This section is our analysis.
- A demonstration on public data with no plan for testing on yours.
- Accuracy figures with no description of how they were measured.
- Contract terms that shrink the supplier's data handling duties to almost nothing.
- No named person on the supplier's side accountable for security and data protection.
What to watch next
Watch PP 33/2026 take effect on 16 January 2027, and watch for the standard contractual clauses the data protection authority is to issue for transfers abroad, which will shape contracts with suppliers that use overseas platforms. Watch whether that authority is formed. And watch whether Komdigi's 2023 ethics circular is followed by binding AI rules; until then, it and OJK's guide for banks remain the published reference points for responsible AI in Indonesia.
Sources
- Perbedaan Pengendali dan Prosesor Data Pribadi Menurut UU PDP · Hukumonline(opens in a new tab)
- PP 33/2026 Mulai Berlaku Januari 2027: Apa yang Wajib Disiapkan Perusahaan dalam Pelindungan Data Pribadi? · SIP Law Firm(opens in a new tab)
- Surat Edaran Menteri Komunikasi dan Informatika Nomor 9 Tahun 2023 tentang Etika Kecerdasan Artifisial · JDIH Komdigi(opens in a new tab)
- Peluncuran Tata Kelola Kecerdasan Artifisial Perbankan Indonesia · Otoritas Jasa Keuangan(opens in a new tab)
- OJK: Bank perlu bentuk komite AI untuk awasi tata kelola · Antara(opens in a new tab)
- Perkuat Transformasi Digital, OJK Terus Kembangkan Tata Kelola AI Perbankan · Infobanknews(opens in a new tab)
- OJK Rilis Panduan Tata Kelola Kecerdasan Artifisial, Dorong Bank Membentuk Komite AI · Kontan(opens in a new tab)
This briefing was prepared by the Indotek editorial desk with AI assistance, from the public sources listed above. It is general information, not legal or professional advice.
AI & Agentic AI
How Indotek can help
The test of an AI partner is whether a system can be measured, controlled and explained before it touches customer data. That is how Indotek approaches its AI work.
- Start from the decision: we define the business decision or task first, then choose the simplest technique that performs it reliably, which is often not the largest model.
- Evaluate before you scale: every AI feature we build ships with an evaluation set, so quality is measured rather than assumed and regressions are caught before users see them.
- Humans stay in command: the AI agents we build act inside explicit permissions, log what they do and hand off to a person when confidence is low or stakes are high.
- Grounded in your own content: our generative AI and intelligent search work from your approved content and respect your existing access controls.



