Nigeria’s AML Arms Race: As Banks Deploy Agentic AI, Criminals Are Learning to Automate the Laundering Game

The next threat to Nigeria’s anti-money laundering regime may not arrive through another anonymous account, shell company or suspicious cash deposit. It may arrive as software capable of thinking through the weaknesses of the financial system, testing them repeatedly and adapting when controls begin to close in. That is the emerging compliance dilemma surrounding agentic artificial intelligence.

Financial institutions are presently deploying a raft of artificial intelligence skills, softwares and agents to identify unusual transactions, screen customers, detect patterns and reduce false positives. The next generation appears different, especially with Agentic AI designed not merely to generate an answer when prompted, but to pursue a task, assemble information, take actions, evaluate results and continue working through a defined objective. That distinction could fundamentally alter the Anti-Money Laundering landscape.

A recent Money Laundering report describes agentic AI as a new and expanding development in financial crime compliance. The technology could automate much of the evidence gathering that currently consumes investigators’ time, allowing compliance professionals to concentrate on higher value analysis and judgement.

For banks, fintechs, payment companies and other regulated institutions in Nigeria, the attraction is obvious.

AML investigations can be painfully fragmented. An investigator may need to examine transaction histories, customer records, account relationships, sanctions information, Know- Your- Customer (KYC) documentation, adverse media and other data sources before determining whether a case warrants escalation.

Agentic AI could potentially connect those systems, assemble an evidence package, identify relevant transaction patterns and produce an investigation narrative before a human investigator reviews the case. That is potentially transformative.

It could also become dangerous if the technology is deployed faster than governance frameworks can keep pace.

The first compliance question should therefore not be whether Nigerian financial institutions can afford agentic AI. It should be whether they can afford to deploy it without creating a new category of operational, regulatory and financial crime risk.

From detection to autonomous investigation…

Traditional transaction monitoring largely operates by identifying activity that matches predefined rules or risk models. A transaction crosses a threshold, deviates from an established customer profile or matches a suspicious pattern, and the system generates an alert.

Generative AI can then help an analyst interpret or summarise the information. Agentic AI potentially goes much further.

The Money Laundering report describes systems capable of gathering transaction histories, querying databases, investigating connected entities, summarising findings, prioritising cases and assisting with suspicious transaction reports. This creates an important shift in responsibility.

The machine is no longer simply identifying something for the compliance officer to investigate. It is participating in the investigation itself. That could dramatically improve efficiency in Nigeria, where financial institutions face large transaction volumes and increasingly complex fraud and money laundering typologies. But efficiency is not the same as effectiveness.

A system that processes millions of transactions faster can also make millions of incorrect assumptions faster. The danger becomes greater where AI-generated conclusions are accepted without adequate human challenge.

The criminal gets the same technology….

This is arguably the most important part of the debate. The AML industry is preparing to use AI to fight financial crime. Criminal networks are not required to wait.

The technology could be used to create convincing documentation, identify weaknesses in due diligence procedures, construct persuasive narratives around suspicious transactions and automate attempts to evade monitoring systems.

The report cites concerns that agentic AI could help criminals structure payments at different times or below reporting thresholds and continuously adjust their methods. It also highlights the possibility of AI being used to generate large volumes of communications designed to persuade financial institutions that suspicious transactions are legitimate. That changes the economics of financial crime.

A sophisticated criminal network traditionally needed people with specialist knowledge of banking procedures, company structures, payment systems and compliance controls.

Agentic systems could potentially reduce that human requirement. A criminal does not necessarily need to understand every weakness in a financial institution’s controls if software can continuously test different approaches and identify which ones are successful.

The implication for Nigeria is significant. The country’s financial system is already dealing with mule accounts, identity fraud, account takeovers, cybercrime, fintech abuse, digital payments and increasingly sophisticated social engineering. AI could make these existing threats more scalable.

The KYC problem becomes harder…

Know Your Customer controls could become one of the principal battlegrounds. Banks and other obliged entities depend on identity documents, corporate records, source of funds information, beneficial ownership information and customer declarations.

But the same technologies that make legitimate digital onboarding faster can make fraudulent identities more convincing.

Synthetic identities could become more sophisticated. Forged documents could become harder to distinguish from genuine documents. Corporate structures could be constructed to obscure beneficial ownership. Fraudsters could potentially use AI to generate explanations designed to make questionable sources of funds appear legitimate.

This means KYC cannot remain a document-checking exercise.

Financial institutions will increasingly need to establish whether the person, business, transaction and economic explanation actually make sense when considered together. That requires contextual intelligence and human judgement.

The human being remains the control…

One of the strongest messages from the emerging agentic AI debate is that human oversight is not becoming obsolete. It is becoming more important.

The Money Laundering report quotes compliance and technology specialists stressing the need for human experts to validate AI conclusions, particularly because erroneous or fabricated reasoning can contaminate an investigation. This has an important regulatory implication.

A compliance officer should not be able to defend a regulatory decision simply by saying, “the AI recommended it”. The regulated institution remains responsible.

If an institution files a suspicious transaction report, freezes an account, rejects a customer or escalates a case, there must be a defensible basis for that decision. AI can support the decision; it cannot become the accountability shield behind it.

The audit trail becomes critical…

For Nigerian regulators, one of the most important requirements for agentic AI should be explainability.

Every significant AI-assisted compliance action should leave a verifiable trail showing what information the system accessed, what instructions it was given, what conclusion it reached, what evidence supported the conclusion and what human intervention occurred before a regulatory decision was made.

The principle should be simple. If an AI system can make or materially influence a compliance decision, the institution must be able to reconstruct what happened.

This is particularly important in investigations that could eventually reach court.

An AI-generated conclusion without a reliable chain of evidence may have little value when challenged.

The technology industry is already recognising this problem. The Money Laundering report highlights the importance of evidence, verification, validation and transparent decision records when agentic AI is used in AML. For Nigerian banks, this should become a procurement requirement rather than an afterthought.

A new third-party risk…

There is another compliance vulnerability that deserves greater attention.

Many financial institutions will not build their own agentic AI systems. They will buy them. That creates third-party risk.

Who owns the underlying model? Where is customer data processed? Can the vendor access transaction information? Is data transferred outside Nigeria? Can the model learn from confidential information? What happens when the vendor updates the system? Can the bank independently audit the model? What happens if the vendor’s system makes a material error? These questions should become part of vendor due diligence.

The traditional approach of assessing a technology provider primarily through cybersecurity and service-level agreements may no longer be sufficient.

Financial institutions will need to assess AI governance, model risk, data governance, access permissions, auditability, resilience and human override mechanisms.

The regulator’s challenge…

For regulators, the emergence of agentic AI creates a difficult balancing act. A regulator that moves too slowly could allow institutions to deploy powerful systems without adequate controls.

A regulator that is too prescriptive could discourage legitimate innovation and leave Nigerian institutions technologically behind their international competitors.

The better approach may be principles-based supervision.

Institutions should be required to demonstrate that AI used in AML is proportionate to risk, properly governed, independently tested, auditable and subject to meaningful human oversight.

The regulator should also be able to ask a straightforward question. If this AI system fails tomorrow, who is accountable? The answer cannot be the algorithm.

The coming AI arms race….

The AML industry is entering an arms race. Financial institutions will use increasingly autonomous systems to find suspicious activity faster, investigate cases more efficiently and reduce the cost of compliance.

Criminals will use increasingly autonomous systems to conceal suspicious activity, manufacture legitimacy and test financial controls.

The advantage will not necessarily go to whoever has the most sophisticated AI. It may go to whoever has the better data, stronger governance and more disciplined human oversight. That is the central lesson for Nigeria.

The report warns that Agentic AI should not be treated simply as another compliance technology upgrade. It represents a change in the relationship between financial crime, technology and control. The banks that deploy it intelligently could reduce investigation backlogs, improve detection and allow compliance officers to concentrate on genuinely complex cases. Those that deploy it carelessly could create new vulnerabilities while believing they have strengthened their defences.

And regulators face an even bigger challenge. They must ensure that the same financial system adopting autonomous machines to detect criminals is also prepared for criminals using autonomous machines to defeat the system.

The future of AML may therefore not be a contest between humans and machines.

It may be a contest between machines designed to protect the financial system and machines designed to exploit it. The decisive factor will remain the human being standing between the two.