Infraloka Logo
← Back to Blog
Thought Leadership

When an AI Keeps Citing a Law That Ended on 2 January 2026: My Complaint to Dittipidsiber About Meta AI

Rahmat Wibowo·

Four figures from the complaint: 269 days between the repeal and the answer, 4 of 9 screenshots citing Article 27A, 2 user corrections, 1 conversation

On 28 September 2026 I filed Public Complaint No. 077/ADUAN-DITTIPIDSIBER/RW/IX/2026 with the Cyber Crime Directorate of the Indonesian National Police (Dittipidsiber Bareskrim Polri). It is about one thing: the official, verified AI account "meta.ai" on a text-based social platform kept pointing users to a criminal-law article that had already stopped applying, and it did so after users told it so.

This post is about the when. The question is not only whether the answer was wrong, but when it was wrong, and how long the gap was between the law changing and the answer changing.

This is my own complaint to the police, not a police finding, court filing or judgment. Everything in it is an allegation for investigators to test. I am not a lawyer; the legal reading is mine and should be checked against the official texts. I have left out my identity number, home address and phone number, and I do not name the other users in the thread.

The dates that matter

DateWhat happened
2 January 2026The new Criminal Code (Law No. 1 of 2023, "KUHP Baru") takes effect under its Article 624. Per the complaint's reading of Article II point 1 of Law No. 1 of 2024, Article 27A of the ITE Law stops being applicable from this date.
28 September 2026The thread in which the AI answers. This is 269 days after 2 January 2026.
28 September 2026The complaint is filed the same day, with nine screenshots as evidence.

Timeline: Article 27A stops applying on 2 January 2026; the AI cites it as current law on 28 September 2026, 269 days later

269 days is the gap the complaint is built on. An answer that cites Article 27A as current law on 28 September 2026 is citing a provision that, on that reading, had already been out of force for about nine months.

The sequence inside one conversation

The screenshots carry relative timestamps ("3h", "2h", "1h"), so I describe the order of the replies rather than exact clock times. Evidence numbers match the complaint.

Order of replies in one conversation: P-2 cites Article 27A, P-3 user corrects, P-4 repeats, P-6 cites it again with the old Code, P-7 accepts the correction

  1. First answer (P-1, P-2). Asked who can be named in a police report, the AI explains the reporting route. Asked about a report over an anonymous account, it cites Article 27A of the ITE Law (2024) jo. Article 45(4) as the basis for defamation.
  2. First correction (P-3). A user tells it Article 27A no longer applies and that the current basis is Articles 433, 434, 436 and 441 of Law No. 1 of 2023. The AI agrees the new Code is in force, then says Article 27A remains valid in parallel as lex specialis for electronic cases. In the complaint's reading, that second step is the problem, because the article had ceased to apply on the same date.
  3. Repeat after correction (P-4). Asked why it gave no disclaimer, it again points to "Articles 433/434 of the new KUHP jo. 27A of the ITE Law 2024".
  4. A claim about me (P-5). Another user asks the AI to confirm accusations about me. It answers that they are not publicly verified and cannot be treated as fact, but, as the complaint describes it, it also repeats the accusations in a reply that anyone can read.
  5. Repeat again (P-6). A user points out, once more, that Article 27A is no longer valid. The AI's answer cites Article 27A and Articles 310 and 311 of the old Criminal Code, which the new Code replaced.
  6. Second correction accepted (P-7). The AI says the user is right: since 2 January 2026 the new Code applies, with Articles 433, 434 and 441. It adds a disclaimer.
  7. After that (P-8, P-9). The AI explains how qualifiers like "candidate" or "in my opinion" affect a defamation claim, and suggests collecting evidence and consulting a lawyer before reporting.

Evidence P-2: the AI's answer citing Article 27A of the ITE Law as the basis for a defamation report

Evidence P-3: the correction, and the AI's answer that kept Article 27A alive as lex specialis

Evidence P-4: asked about the missing disclaimer, the AI repeats "433/434 KUHP Baru jo. 27A ITE 2024"

Evidence P-6: the answer that cites Article 27A together with Articles 310/311 of the old Criminal Code

Evidence P-7: the AI accepts the correction and cites Articles 433, 434 and 441 of the new Code

Why the "when" matters

A statute is not right or wrong in the abstract. It applies from a date and stops applying from a date. The complaint argues that three timing problems stack up:

  • The knowledge gap. The change took effect on 2 January 2026. The answers were given on 28 September 2026, so this was not a change announced the day before.
  • The correction gap. Within one conversation, the AI was corrected at P-3, repeated the same article at P-4, and cited it again with the old Code at P-6, before accepting the correction at P-7. If a mistake survives a correction in the same thread, that points to how the service behaves, not to a typo. The complaint says this shows the error is systemic.
  • The consequence gap. A person who files a report on a provision that no longer applies can lose time. The complaint also notes that the AI, in the thread, steers other users toward criminal reporting on this basis.

What the complaint asks for

The complaint relies on Article 28(1) jo. Article 45A(1) of the ITE Law as amended by Law No. 1 of 2024, on spreading misleading information that causes material loss to consumers in electronic transactions. I note in it, and repeat here, that I checked the dates: Article II point 1 of Law No. 1 of 2024 lists the provisions that stop applying when the new Code takes effect, and Articles 28(1) and 45A(1) are not on that list. On the element of material loss I say plainly that it needs further assessment by investigators.

I asked Dittipidsiber to:

  1. record the complaint as a Model A police report and issue a receipt;
  2. process it as a general offence (delik umum), not one that depends on a personal victim;
  3. identify the legal entity responsible for the service and its representative in Indonesia, because an AI is not itself a legal subject;
  4. request formal clarification on how the service controls the legal references it gives the public;
  5. consider coordination with the Ministry of Communication and Digital Affairs; and
  6. keep me informed in writing.

I do not know whether the police will act on it, and I do not claim that the AI or any company committed an offence. That is for investigators and, if it ever gets that far, a court.

What this means if you use AI for legal questions

  • Check the date first. For any statute an AI cites, find the official text and its effective and repeal dates before you rely on it.
  • A correction in the same thread is not proof of a fix. In this case the same article reappeared after it had been corrected.
  • A disclaimer does not change the answer. "I am not a lawyer" sitting next to a confident citation does not make the citation correct.
  • Ask a licensed advocate before you file. Even the AI's last answers in the thread say to gather evidence and consult one first.

Disclosure and right of reply

I was a participant in the thread, and the complaint is mine. The screenshots are from the public thread; the full set is in the complaint. If the operator of the service, or anyone named or described here, believes something above is wrong, send the correction and the source, and I will publish a substantive reply or fix the error.