Business / Nigeria

Unchecked AI agents threaten banks’ payment systems

Nigerian banks and fintechs must prevent artificial intelligence agents from authorising payments independently, TransFi chief Raj Kamal warns, citing risks of widespread losses and systemic payment disruptions. JUSTICE OKAMGBA writes

Nigerian banks and fintech companies should prevent artificial intelligence agents from granting final payment authorisation as they expand automation, TransFi Chief Executive and Founder Raj Kamal has warned, arguing that a single software error could trigger widespread financial losses and disrupt interconnected payment systems.

Kamal said AI agents should initially be restricted to interpreting instructions and recommending actions, with payment authorisation and settlement governed by deterministic rules and human oversight where necessary.

Unlike conventional software that follows predefined instructions, agentic AI can interpret requests, make decisions and execute multi-step tasks through connected systems. In financial services, this could enable an agent to initiate transactions independently, creating risks if it misinterprets a customer’s instructions or repeats an error across multiple accounts.

“The IMF’s three-layer model confines AI to interpreting intent and keeps authorisation and settlement deterministic,” Kamal said in an interview with The PUNCH. He recommended that Nigerian financial institutions begin with narrowly defined applications in which mistakes are inexpensive and easily reversible.

His warning comes as international banks tighten controls on external AI agents seeking access to customer accounts and the ability to execute transactions.

JPMorgan Chase, the largest US bank by assets, is blocking third-party AI agents from logging into retail customers’ accounts using their banking credentials and performing actions such as transferring money or paying bills. It permits limited, customer-authorised data sharing through approved providers, allowing access to information such as account balances without granting transaction privileges.

Chief executive Jamie Dimon warned in October that allowing unrestricted external agents into the bank’s systems would expose customers and the institution to unacceptable risks. He also cautioned that customers who provide their credentials to external agents could face a transfer of liability if the tools act against their instructions.

Kamal’s concerns extend beyond conventional cybersecurity breaches, which typically involve unauthorised access to systems. An AI agent operating with legitimate access could make an incorrect decision that passes through existing controls because the transaction appears valid.

The risk is compounded by the prospect of multiple financial institutions relying on a small number of AI models and technology providers. A shared weakness could affect several institutions simultaneously, while autonomous systems could repeat erroneous transactions thousands of times before staff detect the problem.

Nigeria’s payment infrastructure has already experienced costly failures involving conventional software.

In September 2024, a logic error in the Nigeria Inter-Bank Settlement System’s NIP platform reportedly credited N13.66bn to 176 accounts across 19 banks and fintech companies without debiting the originating accounts. By the time the error was detected three days later, N4.19bn remained unrecovered, according to Kamal.

The structure of Nigeria’s instant transfer infrastructure presents an additional challenge. Although customers’ accounts can be credited immediately, interbank settlement takes place in batches. A surge of erroneous transactions could therefore create obligations faster than the settlement process could absorb them.

The effects could spread beyond the institution where a failure originated. Banks receiving suspicious inflows might restrict transactions or hold back funds to protect themselves, disrupting payments for customers and businesses unconnected to the original error.

Kamal cited another incident in which 718 customers transferred N1.13bn from a Nigerian bank into accounts at 45 banks and fintech companies following a glitch in July 2024. The affected bank subsequently sought to recover the money through the courts.

He warned that AI-driven failures could amplify such incidents, leaving customers unable to receive salaries and merchants struggling to confirm payments. Previous outages affecting NIBSS, which operates critical national payment infrastructure, have also highlighted the economy’s dependence on interconnected digital payment services.

The question of who bears the cost of such failures is another concern. Kamal argued that the bank or fintech deploying an AI agent should take initial responsibility for compensating affected customers, rather than requiring them to establish whether the institution, software developer or technology provider caused the loss.

“My view is that the institution deploying the agent and owning the customer relationship carries first-line liability, which in most cases means the bank or fintech,” he said. “A customer should never have to work out whose model failed before getting their money back.”

Financial institutions could subsequently seek compensation from technology vendors under contractual arrangements. But Kamal warned that existing agreements often limit vendors’ liability to fees paid over a specified period, potentially leaving banks exposed to losses far exceeding the compensation available.

Insurance could provide another layer of protection, although existing cyber policies may not cover losses caused by an institution’s own AI agent acting autonomously. Kamal also pointed to Nigeria’s low insurance penetration, which he put at about 0.5 per cent of gross domestic product, as a constraint on the domestic market’s capacity to absorb emerging risks.

He called on the CBN to introduce safeguards before autonomous AI becomes more deeply embedded in the country’s payment infrastructure.

His recommendations include requiring the new National Payment Stack to identify agent-initiated transactions, imposing transaction-velocity limits that can pause suspicious activity without shutting down the entire payment system, and establishing a register of AI agents authorised to initiate payments.

He also advocated pooled incident reporting through NIBSS, common standards for technology vendors and regularly tested emergency shutdown mechanisms. Such measures would be particularly important for smaller institutions that cannot afford extensive in-house AI risk-management teams.

Distinguishing between AI chatbots and autonomous agents will be important as Nigerian financial institutions expand their use of the technology. A chatbot primarily answers questions or routes customer requests, whereas an agent can execute multi-step tasks independently using connected tools and systems.

The presence of a customer-service chatbot does not, therefore, necessarily mean a bank has given AI independent authority to transact. The critical question is whether a system can initiate or execute financial actions without a separate, rule-based authorisation step.

Kamal’s argument is not that banks should abandon AI adoption, but that automation must not weaken the controls protecting customers’ money.

Nigerian institutions processed about 11 billion transfers worth N1.07qn in 2024, he noted, arguing that the scale of the country’s digital payments market made preparation urgent.

He urged banks to use the period before the CBN’s AI-related anti-money laundering compliance deadlines in 2027 and 2028 to establish controls, test failure scenarios and ensure that autonomous systems cannot turn a single mistake into a wider financial disruption.

Share this story WhatsApp Facebook X
Originally published by Punch
View on Punch

Comments 0

No comments yet. Start the conversation.

Loading the next story…