AI and IoT Projects Now Need a DPIA in Indonesia: What PP 33/2026 Asks Before Go-Live
November 13, 2026 · 6 min read · Etzal Group Team
Indonesia's Government Regulation 33 of 2026 (PP 33/2026), which implements the Personal Data Protection Law and takes effect on 16 January 2027, requires a data protection impact assessment (DPIA) before high-risk processing begins. According to RSM Indonesia's summary (18 September 2026), the official elucidation names AI, machine learning, smart technology and IoT as examples of new technology.
For companies building a chatbot, an automated scoring model, CCTV analytics or a sensor network, this changes the project schedule. The DPIA is no longer a document written after the system is live. It is a condition before personal data starts to flow.
A note on sources: every article number here comes from law-firm and consultancy summaries, not from the official text.
What the regulation asks for
According to RSM Indonesia's summary:
- Article 120 paragraph (2) lists seven categories of high-risk processing that require a DPIA. The elucidation of letter f names AI, machine learning, smart technology and IoT as new technology.
- The DPIA must be completed before processing starts and reviewed when the risk changes (Article 121 paragraphs (1) and (5)).
- Data subjects may object to decisions based solely on automated processing, including profiling, that have legal consequences or a significant impact (Articles 93 to 96).
DFDL's summary (page updated 28 September 2026) lists the triggers: automated decision-making, large-scale specific (sensitive) data, systematic monitoring, new technology, cross-border transfer, processing that may restrict rights, and a combination of risk factors. DFDL also notes that there is no numeric threshold for "large scale".
Veritask (3 September 2026) adds that the DPIA must be documented, and that the advice of the data protection officer (PPDP in Indonesian) must be considered and recorded (Articles 120 to 122).
Two sources differ on cross-border transfers
DFDL places cross-border transfer in its list of DPIA triggers. RSM Indonesia describes third-party data sharing and international transfer as screening indicators, not triggers under the article's wording. We cannot settle which is right without the official text.
The practical stance: if your project sends personal data to a service abroad, such as an AI model API or a foreign cloud region, treat that as a strong reason to run a DPIA until your legal counsel confirms the wording.
There is also no official DPIA format yet, because the supervisory agency has not been established. Check the latest official announcement before you lock a template.
Which projects need a DPIA
The table below is our practical assessment, not a classification from the regulation. Use it as a first screen.
| Project | Risk signal | Suggestion |
|---|---|---|
| AI customer service chatbot | New technology, conversations contain personal data | Run a DPIA |
| Automated scoring (credit, hiring, per-person pricing) | Automated decision, may restrict rights | Run a DPIA, prepare a human review path |
| CCTV with face or behaviour analytics | Systematic monitoring, new technology | Run a DPIA |
| Sensors that record people (attendance, worker location, driven vehicles) | IoT, monitoring | Run a DPIA |
| Pure machine sensors (temperature, vibration, pressure) | No data about people | Record the reasoning, a DPIA is probably not needed |
Be careful with the last row. Many "machine only" sensor systems turn out to store operator IDs, shift schedules or camera footage of the production line. Once data can be linked to a person, the assessment changes.
Because there is no numeric threshold for "large scale", do not argue about the number for long. Write down how many data subjects are involved, why you consider that large or not, and keep the note.
Where it sits in the project schedule
Per RSM's summary, the DPIA is finished before processing starts. In project practice, "processing starts" earlier than people think. A pilot with real customer data is already processing. So is training a model on transaction history.
The order we suggest for a DPIA for AI projects and IoT projects alike:
- Idea stage. Fill in a one-page screen: what data, whose, for what decision, sent where.
- Design stage. Do the full DPIA before a vendor is chosen, because the result shapes contract terms and architecture.
- Before a pilot with real data. The DPIA is signed. If not, the pilot uses synthetic or anonymised data.
- Before production. Check that the promised mitigations are in place.
- After go-live. Review when the risk changes: a new model, added data sources, a wider purpose, a different vendor.
The trade-off is plain. A DPIA at design stage adds one to two weeks up front. A DPIA after the build can force an architectural rework, which costs far more.
What a useful DPIA contains
Without an official format, write a document that answers five things:
- a description of the processing: data, sources, flows, recipients, retention
- why the processing is needed and whether a lighter approach exists
- risks to the people whose data is processed, not only risks to the company
- mitigations, each with an owner and a date
- residual risk and the decision: proceed, proceed with conditions, or stop
For systems that make automated decisions, add one section: how a person objects and which human reviews it. This matches the right to object to automated decisions described in RSM's summary.
Who signs
The rules summarised by Veritask require the PPDP's advice to be considered and recorded. A sensible split of roles follows:
- The business owner of the project signs as the accountable person. Not the vendor, and not IT alone.
- The PPDP or data protection lead gives written advice. If it is not followed, the reason is recorded.
- Directors approve when the residual risk remains high.
- The vendor supplies technical information but does not assess itself.
What to do this week
- List every AI and IoT project that is running or will start before mid-2027.
- Run the one-page screen for each.
- Flag projects whose pilot falls after 16 January 2027 and put the DPIA into their schedule.
- Ask vendors for a description of data flows and processing locations.
Our AI consulting and IoT and control systems teams can help map a project's data flows before its design is locked.
This article is not legal advice. Confirm your project's obligations with legal counsel and the official text of PP 33/2026.
Sources
- RSM Indonesia, 18 September 2026: summary of PP 33/2026
- DFDL, page updated 28 September 2026: Indonesia personal data protection law, GR 33/2026
- Veritask, 3 September 2026: technical rules for personal data protection under PP 33/2026
Frequently asked questions
Does every AI project in Indonesia need a DPIA?
Not automatically. According to summaries by RSM Indonesia and DFDL, a DPIA is required for high-risk processing, and new technology such as AI and IoT is one of the categories when personal data is processed.
When must the DPIA be finished?
According to RSM Indonesia's summary, it must be completed before processing starts and reviewed when the risk changes.
Is there an official DPIA format?
Not yet, because the supervisory agency has not been established. Check the latest official announcement before fixing an internal template.
Related service
WhatsApp AI Chatbot
We build, train, and run a WhatsApp AI agent that answers customers 24/7, qualifies leads, and follows up.
