Skip to main content
Indotek Digital Nusantara

AI & Its Impact

What Is Know Your Agent? Komdigi's Agentic AI Warning for Finance

Komdigi's deputy minister has told companies that knowing the customer is no longer enough: they must also know the AI agents acting on their systems. What Know Your Agent means, why agents that can transact change the risk in finance, the rules that exist so far, and what organisations can prepare while new rules are drafted.

The headquarters of the Ministry of Communication and Digital Affairs on Jalan Medan Merdeka Barat in Jakarta
The Ministry of Communication and Digital Affairs (Komdigi) in Jakarta. Its deputy minister has urged companies to know the AI agents that act on their systems. Photo: Kementerian Komunikasi dan Digital Republik Indonesia, Public domain (opens in a new tab)

What is Know Your Agent, and why is Indonesia's digital ministry talking about it? Over the past three months, Nezar Patria, Deputy Minister of Communication and Digital Affairs (Komdigi), has made the same point in several settings: AI systems are moving from suggesting decisions to taking them, including financial transactions, and companies need to know which AI agents are acting on their systems and what those agents are allowed to do. Last week he added that AI agents are also becoming a tool for cyber attacks. None of this is yet a binding rule, but it shows where Indonesian AI governance is heading, and the financial sector is at the front of it.

What Komdigi means by Know Your Agent

According to a Komdigi press release published on InfoPublik on 16 October 2026, Nezar spoke at a Hukumonline executive breakfast in Central Jakarta on 15 October. He said companies increasingly use agentic AI for customer service, transactions and other business processes, and that it is no longer enough to know the customer, the familiar Know Your Customer duty. Companies should also know the AI agents that interact with their customers or their business systems.

In the release, Nezar said that once machines can make their own decisions, risk rises, so a company needs to understand each agent's identity, its ownership and the limits of its authority. He warned that because agents can act, they may bypass rules, regulations and a company's internal limits, and that a company's counterparty in future may itself be another AI agent, or a person who owns one. He also said data leaks can come from how a new technology is given access and authority to process data, not only from weak system security, and urged security by design and privacy by design from the first design stage.

Why agents that can transact change the risk

Nezar developed the financial angle in August. Katadata reported on 28 August 2026 that he said agentic AI in financial services can now carry out decisions, including transactions, and even transactions in which two AI systems communicate and agree a deal between themselves. A report carried by RCTI+ from IDXChannel the same day attributes the remarks to OJK Digination Day 2026 in South Jakarta on 27 August, while Katadata describes them as a written statement; the two reports agree on what was said but not on the setting.

His reasoning, in both reports, is that the risk differs from AI used as a tool. When AI only recommends, a person still makes the final decision. When a system can decide and act, its errors reach transactions and consumers directly. According to Katadata, he said governance should be prepared before highly autonomous AI spreads in finance, and listed what rules need to settle:

  • the limits of an AI system's authority;
  • human oversight of what it does;
  • who is liable for losses caused by a system's decision;
  • data security and risk management;
  • how to mitigate risk when one AI system interacts with another.

Agents as an attack tool

The second strand is security. Investortrust reported on 11 November 2026 that Nezar said in Jakarta on 10 November that Komdigi is coordinating with the National Cyber and Crypto Agency (BSSN) because AI agents and agentic AI are becoming a new global pattern of cyber attack, and that groups of agents may be used to launch coordinated attacks. He noted that operational protection of critical infrastructure is BSSN's responsibility. The same report cited FortiGuard Labs as detecting about 1.6 million attack activities in Indonesia in 24 hours on 9 November, and a BSSN spokesperson, named only as Prabaswari, as saying BSSN has formed 151 cyber incident response teams, especially in the financial sector, issued 208 security advisories during 2026 and sent 202 threat notifications to financial-sector teams.

On 13 November, JawaPos reported, Nezar said Komdigi is preparing risk-based digital supervision, in which protection is proportional to a technology's potential impact on people's safety and rights, with finance, health and education among the sectors needing particular attention. He said AI makes cybercrime easier to commit and harder to recognise, citing phishing, data theft and fraud with fake voices and faces.

What rules exist, and what is still coming

Know Your Agent has so far appeared in a minister's remarks and a press release, not in a regulation, and no source we read gives a timetable for turning it into one. The existing reference point, cited by both August reports, is Circular Letter of the Minister of Communication and Informatics Number 9 of 2023 on AI ethics. Nezar said a National AI Roadmap and AI ethics guidance are being processed to be issued together as a presidential regulation. Separately, Infobanknews reported on 19 August that he said Komdigi is preparing a bill on cyber security and resilience to strengthen defences against AI-based attacks. He has also called for Komdigi and OJK to align AI governance principles and to test innovations in cross-sector sandboxes.

In our reading, that leaves financial firms in a familiar position: the direction is clear, the detail is not. The points Nezar keeps returning to, namely identity, ownership, limits of authority, human oversight and accountability, are the questions any supervisor is likely to ask about an agent, whatever form the final rules take.

What organisations can do now

The steps below are our analysis of sensible preparation, not a statement of what the law requires.

  • Keep an inventory of every AI agent in use, including those inside vendor products, with a named owner for each.
  • Write down what each agent may do: which systems it can reach, which actions it can take and any value limits on transactions.
  • Require human approval for high-stakes or irreversible actions, and define when an agent must hand a case to a person.
  • Log every action an agent takes, with enough detail to reconstruct who or what authorised it.
  • Decide how your systems will recognise and treat requests that come from someone else's AI agent rather than a person.
  • Test agents against misuse and manipulation before release, and keep testing after they are live.

What to watch next

Watch for the presidential regulation on the National AI Roadmap and AI ethics, and whether it mentions AI agents or Know Your Agent explicitly; any joint guidance or sandbox from Komdigi and OJK on AI in finance; progress on the cyber security and resilience bill; and further BSSN advisories on AI-based attacks against the financial sector.

Sources

  1. [Siaran Pers] Tekankan KYA, Wamen Nezar Patria: Bukan Lagi Cukup Kenali Pelanggan · InfoPublik (Komdigi press release)(opens in a new tab)
  2. Agentic AI Masuk Sektor Keuangan, Komdigi Soroti Risiko Transaksi Tanpa Manusia · Katadata(opens in a new tab)
  3. Agentic AI Mulai Bisa Eksekusi Transaksi Keuangan, Komdigi Siapkan Regulasi untuk Antisipasi Risiko · RCTI+ (IDXChannel)(opens in a new tab)
  4. Kemenkomdigi Waspadai Serangan Siber Berbasis Agentic AI · Investortrust(opens in a new tab)
  5. Komdigi Siapkan Pengawasan Digital Berbasis Risiko, Keuangan hingga Kesehatan jadi Prioritas · JawaPos(opens in a new tab)
  6. Kerugian Fraud AI Tembus Rp9,1 Triliun, Komdigi Soroti Pentingnya Pertahanan Siber · Infobanknews(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

Knowing an agent's identity, owner and limits is easiest when those limits are built into the agent from the start rather than added after an incident. Designing AI agents that act through governed access, with approval points and records, is the AI work Indotek does.

  • AI agents: we build agents that plan and carry out multi-step work across systems through governed tool access, with approval points where human judgement matters, such as before a transaction is executed.
  • Humans in command: our agents act inside explicit permissions, log what they do and hand off to a person when confidence is low or the stakes are high, so every action can be traced to who or what authorised it.
  • Evaluation before scale: every AI feature ships with an evaluation set, so an agent's behaviour, including how it responds to misuse, is measured before customers rely on it.
  • Workflow automation: we combine deterministic rules, such as transaction limits, with AI judgement, so exceptions are routed to people rather than dropped or decided silently.

Let's build what comes next

Whether you are modernising a legacy environment, building an AI-powered product or exploring a new digital business, let's create technology that delivers lasting value.