PP 33/2026 and AI: what to have ready by 16 January 2027
PP 33/2026 is the implementing regulation of Indonesia’s Personal Data Protection Law (UU PDP, Law 27 of 2022), and it applies from 16 January 2027. It does not ban AI. But when AI reads personal data, such as customers’ names and phone numbers or staff records, a company needs a few things in place first.
Biissa team · Updated 16 November 2026
In short
- Full name: Peraturan Pemerintah Nomor 33 Tahun 2026, the implementing regulation of UU PDP (Law 27 of 2022).
- Signed on 16 July 2026. It applies from 16 January 2027, six months after it was promulgated (Article 225).
- Sanctions include a written warning, a stop to the processing, an order to delete data, and a fine of up to 2% of yearly revenue (Articles 184 and 185).
- The supervisory agency it names had not been set up when this guide was written. Until its rules exist, processing may continue as long as it does not conflict with PP 33/2026 (Article 223).
Who is who when you use AI
Your company decides why and how personal data is used, so it is the controller (pengendali data pribadi). A company that processes the data for you, such as an AI service or the team that builds AI into your work, is a processor (prosesor data pribadi). The company that runs the AI model behind that service processes the data too, as a further processor.
Each of them needs to be covered in writing, and the controller has to supervise its processors (Articles 126 and 127).
What to have ready
A legal basis for each use
Before processing, name the basis for each purpose, such as consent, a contract or legitimate interest. Legitimate interest needs a written analysis.
Articles 30, 53 and 54
A written agreement with every processor
It must cover the scope, method and purpose, the types of data and the people concerned, the duration, each side’s rights and duties, oversight and audit, how disputes are settled, further processors, and a contact. A provider’s standard terms may not cover all of it, so check them.
Articles 13 to 15
An impact assessment before you start
Some processing carries a high risk and needs an assessment first. New technology is one case, and the explanatory note to the article says it includes AI. Scoring people and matching data sets are others.
Articles 120 and 121
A record of processing
Write down what data you process, why, who receives it, where it goes and how long you keep it. Each processor keeps its own record too.
Articles 74 and 75
Security measures that fit the risk
Such as pseudonymisation or encryption, the ability to recover, and regular testing.
Article 123
A plan for data going abroad
Many AI models run outside Indonesia. A transfer abroad needs a mapped and assessed flow, notice to the people concerned, and one of the grounds the regulation lists. The country list and the standard contract clauses are for the agency to issue, and consent covers only transfers that are not repeated and involve few people.
Articles 160 to 173
A breach plan
If personal data leaks, tell the people concerned and the agency within 72 hours. A processor tells you first.
Articles 114 to 118
A way to answer people’s requests
People can ask to see, correct or delete their data, and each request has a 72-hour deadline.
Articles 71, 76 to 78 and 80 to 87
What this means for an AI project
- Give the AI only the data the job needs (Article 59).
- Replace names and numbers with codes before the model reads them. This is pseudonymisation, a security measure under Article 123. Codes can be turned back into names, so the data is still personal data and the rest of this list still applies.
- Keep a person’s approval before anything is sent or recorded, and log each step.
- Know which companies process the data, and in which countries, and have each one in writing.
- Do the impact assessment before the pilot, not after it.
How Sambung handles it
- Recorded names and numbers are replaced with codes before the AI reads them, and every step is logged.
- A person on your team approves every action before anything is sent or recorded.
- Your director signs a data processing agreement before we read any customer chat.
- You approve every outside party that processes your data, and where, before we start.
- Your data is not used to train AI.
What is still open
- The supervisory agency has not been set up yet, so its technical rules, its country list and its standard contract clauses do not exist.
- Until they do, Article 223 lets processing continue as long as it does not conflict with PP 33/2026.
- Some questions turn on the facts of each company. Have a lawyer check your own case.
Common questions
Does PP 33/2026 ban using AI on personal data?
No. It sets conditions: a legal basis, written agreements with processors, an impact assessment for new technology such as AI, security measures, and rules for data going abroad.
When does PP 33/2026 apply?
From 16 January 2027. UU PDP, the law it implements, has applied in full since 17 October 2024.
Is replacing names with codes enough?
No. It lowers the risk and counts as a security measure, but coded data can be linked back to people, so it is still personal data. The legal basis, the agreements, the assessment and the records are still needed.
Who is the controller when we use an AI service?
Your company, because it decides why the data is used. The AI service is a processor and needs a written agreement with you.
What are the penalties?
Administrative sanctions: a written warning, a stop to the processing, an order to delete data, and a fine of up to 2% of yearly revenue, weighed by the harm, how long it lasted, how many people were affected and other factors (Articles 184 and 185).
Sources
General information, not legal advice. The article numbers are from the published text of PP 33/2026. Have a lawyer check your own case.