On 31 July 2026, OpenAI disclosed that it had disrupted a Cambodia-based criminal operation using ChatGPT to support investment, romance, gambling, and law-enforcement impersonation scams. The suspected network created false identities, translated conversations, generated images of forged documents, and adapted its messages to different victims. Some activity also suggested links to the wider ecosystem of scam compounds, human trafficking, and forced criminality.
The case exposed more than another misuse of generative AI. It revealed an institutional weakness. A suspected operation in Cambodia interacted with global technology platforms, approached victims across borders, and sought payments that could pass through financial intermediaries in several jurisdictions. No government could see the entire chain.
Criminal networks increasingly operate across a region as one system. States still respond nationally as separate jurisdictions.
Agentic financial fraud is therefore becoming a test of state capacity. AI-enabled criminal networks can move across platforms, payment systems, and borders, while governments remain constrained by territorial authority and different legal procedures. For ASEAN and other fast-digitizing regions, the strategic question is whether payment integration can be matched by regional enforcement, intelligence sharing, and victim protection.
Stay ahead of the geopolitical week.
MD Briefing delivers expert analysis across five global fronts — the Indo-Pacific, energy, geoeconomics, European security, and the Middle East — every Monday morning. Free.
When persuasion becomes a system
The Cambodian operation was not fully autonomous. Humans directed ChatGPT to perform particular tasks, including persona creation, translation, target research, and emotional manipulation. AI supported the operation without independently controlling the campaign.
The direction of travel is nevertheless visible. Agentic systems could connect these tasks, retain context across conversations, and pursue a financial objective with less human supervision. An agent might identify potential victims, select an appropriate persona, maintain contact, and adapt its story before deciding when to request money.
In its Global Financial Fraud Threat Assessment, published on 16 March 2026, INTERPOL warned that agentic AI could autonomously plan and execute complete fraud campaigns. The assessment described how such systems could support reconnaissance, credential harvesting, system infiltration, and psychologically tailored ransom demands.
The report also cited a Chainalysis analysis of cryptocurrency scams. Chainalysis found that scams with identifiable on-chain links to AI vendors generated an average of US$3.2 million per operation, compared with US$719,000 for scams without those links—around 4.5 times more revenue per scam.
This finding should not be read as a measure of every form of financial fraud. Its importance lies in what it reveals about the changing economics of persuasion. AI can reduce the cost of language, experimentation, and sustained engagement. Criminal groups can offer each victim an apparently personal relationship while managing the operation at an industrial scale.
Financial institutions can establish who authorized a transaction but are less equipped to identify how that decision was shaped. This distinction matters because many investment and romance scams do not begin with account compromise. The victim deliberately approves the payment after a prolonged process of manipulation.
Criminal networks exploit divided sovereignty.
Modern scam operations increasingly resemble distributed production networks. A victim may live in Indonesia. A trafficked worker in Cambodia may maintain the conversation. The false identity operates through an international platform, while the money passes through mule accounts before reaching an exchange or another intermediary. The companies, servers, and financial channels involved may fall under different legal regimes.
A bank may detect an unauthorized transfer, a telecommunications company may identify the number and device, and a platform may hold the messages that preceded the payment. Police in the victim’s country receive the complaint, while authorities elsewhere may control access to the suspects, accounts, or evidence.
AI allows these operations to expand, while jurisdictional fragmentation makes them harder to disrupt. Accounts and identities can be replaced faster than public agencies can assemble evidence and secure cross-border action.
The problem is diplomatic as well as technical. The victim’s government may seek investigation and restitution, but another state controls physical access to the suspected operators. Additional jurisdictions may oversee the platforms or intermediaries used to move the proceeds. Cooperation depends on legal trust, diplomatic relations, and whether governments share the same capacity and willingness to act.
The human dimension further complicates the response. INTERPOL describes financial fraud as increasingly intertwined with organized crime, cybercrime, and human trafficking. The people sending deceptive messages may themselves be operating under coercion. A fraudulent conversation on a phone can eventually involve financial intelligence, labor protection, border control, and relations between states.
Payments are integrating faster than protection.
Emerging economies have strong reasons to connect their payment systems. Faster and cheaper transactions can support regional trade, financial inclusion, and remittances. ASEAN’s cross-border QR initiatives demonstrate the public value that state-supported digital infrastructure can create.
The surrounding protection layer is less developed. A February 2026 report prepared by members of Task Team 2 of the Cross-border Payments Interoperability and Extension—or PIE—task force identified fragmented transaction data, limited automation in pre-transaction checks, immature beneficiary and payment-route pre-validation, and the lack of interoperable real-time fraud registries as major obstacles to fraud prevention. The paper was published on the BIS website as part of the PIE task force’s contribution to the G20 cross-border payments agenda, although its conclusions do not represent the formal views of the BIS, CPMI, or the task force as a whole.
Different privacy laws and regulatory frameworks make it difficult for payment providers to share information across borders. Fraudsters and mule accounts can therefore move between systems, while the recovery of funds becomes slower and more difficult.
Beneficiary verification can confirm the destination of a payment but says little about how the payer reached the decision. It cannot reveal that someone has spent several weeks developing trust in a fabricated identity.
For the Global South, this produces a double asymmetry. Criminal networks can use AI to multiply their reach against individuals who experience every interaction as personal. They can also shift across accounts and jurisdictions, while investigators remain bound by national procedures. Rapid adoption of instant payments often coexists with uneven digital forensics, consumer protection, and asset-recovery capacity.
Cross-border payments may now be completed almost immediately, but sharing warnings, evidence, and legal authority remains much slower.
ASEAN’s next integration test
ASEAN is not starting from zero. Its member states have established a working group on anti-online scams and continued operationalizing the ASEAN Regional Computer Emergency Response Team.
The ASEAN Guide on Anti-Scam Policies and Best Practices was endorsed at the 6th ASEAN Digital Ministers’ Meeting in January 2026. ASEAN foreign ministers then welcomed it in July as the region’s first comprehensive framework for combating scams over telecommunications channels.
These are valuable foundations. The regional architecture, however, remains more developed around telecommunications, cyber incidents, and the exchange of good practices than around the full financial and legal chain of agentic fraud.
National responses are also moving in different directions.
Singapore’s Shared Responsibility Framework took effect on 16 December 2024. It uses a waterfall approach for a defined category of unauthorized phishing transactions. Financial institutions are assessed first, followed by telecommunications providers. Consumers bear the loss when both have fulfilled their prescribed duties. The framework’s scope does not include payments that victims deliberately authorize in investment or romance scams.
Thailand adopted a broader model through its amended Emergency Decree on Measures for the Prevention and Suppression of Technological Crimes. The amendment was published on 12 April 2025 and took effect the following day. It introduced shared liability among financial institutions, covered business operators, and related service providers, while allowing them to avoid liability by demonstrating compliance with standards established by their regulators. The Bank of Thailand subsequently developed sectoral standards to implement this principle for financial institutions and payment providers.
The Philippines chose another route through its 2024 Anti-Financial Account Scamming Act. The law criminalizes money laundering and specified social-engineering schemes, authorizes temporary holds on disputed funds, and can require restitution when financial institutions fail to apply adequate controls or fulfill their holding obligations. Bangko Sentral ng Pilipinas Circular No. 1215 later established implementing requirements for temporary holds and coordinated verification.
These measures show that ASEAN states are acting, but they are not converging on the same allocation of responsibility. Each regime was designed around a different combination of phishing, unauthorized transactions, social engineering, and mule accounts.
The harder case is a payment that a victim knowingly approves after weeks of adaptive, AI-enabled persuasion. ASEAN has no common approach for determining when such a payment should be interrupted, which institution should bear the loss, or how the relevant evidence should move across borders.
This regulatory patchwork reflects how states exercise authority differently: who may suspend a transaction, which institution must compensate a victim, what evidence can be shared, and when private actors are protected for intervening. Once payments cross borders, those national choices begin to affect one another.
Turning geopolitical awareness into capacity
In recent discussions with senior leaders from banks and financial regulatory institutions in Indonesia, I found that concern had begun moving beyond the detection of individual scams. The discussion increasingly focused on how AI-enabled risk travels through payment partners, technology providers, and jurisdictions—and which authority can intervene when the relevant signals are divided among them.
These conversations provide a field signal whose regional reach remains uncertain. Awareness is emerging at the top, while the mechanisms for translating it into cross-border action remain uneven.
States must set the legal architecture. They determine the basis for information sharing, temporary payment holds, evidence preservation, liability, and asset recovery. ASEAN institutions can make these national arrangements more compatible. Banks, telcos, platforms, exchanges, and AI providers can then supply the operational signals required to detect and interrupt fraud.
This does not require a supranational enforcement body, which would sit uneasily with ASEAN’s institutional structure. It requires practical agreements on response times, evidence standards, temporary freezes, and the treatment of victims across jurisdictions.
Greater visibility also carries risk. A regional anti-fraud system with broad access to transactions, communications, and behavioral data could become an infrastructure of surveillance. Governments need sufficient shared visibility to disrupt criminal networks without subjecting every citizen to permanent suspicion. Liability rules must also protect victims without encouraging excessive monitoring or indiscriminate transaction blocking.
Navigating these trade-offs is becoming part of state capacity.
ASEAN’s payment integration will remain strategically incomplete if money can move regionally while warnings, evidence, and legal authority remain confined within national borders. Agentic fraud raises a larger issue than whether banks can detect suspicious transactions. It will reveal whether states can protect citizens when criminal networks learn to operate across the region faster than governments can act together.
