Track · AI in Product5 min read
CBN warns Nigerian banks: AI agents widen cyberattack risk
Nigeria’s central bank urges banks to bound AI agents’ access after citing an Anthropic report that AI handled 80–90% of a large cyberattack targeting roughly 30 organizations.

Context
At ISSAN’s 2026 conference, the CBN said AI access must not exceed what institutions can safely manage.
The CBN cited an Anthropic investigation in which AI reportedly handled 80–90% of an attack on roughly 30 organizations.
Nigeria processed about 47.88 billion electronic payments worth N3.46 quadrillion in 2025.
Electronic-payment fraud losses fell from N52.26 billion in 2024 to N25.85 billion in 2025, a drop of more than 50%.
The CBN’s fraud scenario used an urgent WhatsApp and telephone request for N50 million.
By Daniel Okafor
The Central Bank of Nigeria (CBN) has told banks not to give AI agents more access, authority or autonomy than they can safely manage.
At the 2026 Annual Cybersecurity Conference of the Information Security Society of Africa-Nigeria (ISSAN), Rakiya Yusuf, director of the CBN’s Payments System Supervision Department, put it directly: “For safety and resilience, AI should never be given more access, authority, or autonomy than the implementing institutions can safely manage.”
What is the CBN warning about?
Her concern targets deployments linked to customer records, institutional infrastructure and internal data systems. Without safeguards and access controls, those applications can become entry points for attacks; without firm boundaries, compromised credentials, malicious prompts or human errors can multiply the damage.
Yusuf also urged banks to adopt security by design—building controls into systems from the start—tighten identity and access controls, share threat intelligence, test resilience and maintain human oversight. They must secure AI supply chains, protect institutional data and ensure employees and managers understand the technology before deployment.
The risk is not confined to criminal outsiders. Poorly governed systems can create internal threats, and a legitimate organization’s implementation choices can become the weak point. The practical rule is simple: AI access must match the institution’s ability to supervise it.
How is AI changing the attack?
AI can strengthen Nigerian banks as well. Yusuf cited uses in transaction monitoring, detection of unusual customer behavior and fraud patterns, customer authentication and anti-money laundering controls. Regulators can also analyze large transaction datasets more effectively through supervisory technology, or suptech.
The same scale helps attackers. Blaise Ijebor, director of the CBN’s Risk Management Department, said AI lowers the entry barrier for cybercrime by letting people with limited technical skills generate convincing corporate emails, mimic writing and voices, create synthetic identities, write malicious code and automate attack stages.
He called AI potentially the greatest productivity tool ever handed to cybercriminals, giving attackers the same speed, scale and sophistication that businesses seek through digital transformation.
At the conference, Ijebor cited an Anthropic investigation in which AI reportedly executed 80% to 90% of a large-scale attack against roughly 30 technology companies, financial institutions and government agencies. He framed this as a shift from criminals using AI for assistance to people directing agents that independently complete most of an attack.
That matters more in Nigeria’s rapidly expanding digital financial ecosystem, where instant payments can move stolen funds just as quickly. Ijebor warned that criminals can combine AI-generated deception, information gathered from social media, stolen credentials and infrastructure weaknesses to trick employees into authorizing payments. His example: an employee receives a WhatsApp message and telephone call, apparently from the chief executive, demanding an urgent N50 million transfer.
Voice cloning and synthetic video weaken checks based on appearance, voice or writing style. “Instant payment can also mean instant losses,” Ijebor said. Authentication must answer a harder question: not whether a message looks genuine, but whether the bank can independently prove it is.
The defensible control is independent verification, not sharper intuition about whether a familiar voice or message seems real.
What should banks do before and after an incident?
Urgency, confidentiality and instructions not to call back must never override established authentication controls. Ijebor urged banks to strengthen identity verification because visual, vocal and stylistic resemblance can no longer establish that a request is genuine.
Containment matters as much as prevention. Ijebor said banks need the capacity to keep critical services running, recover from breaches and preserve customer confidence. That means regularly testing backups, running cyberattack simulations and assessing vulnerabilities across technology suppliers; prevention alone cannot guarantee continuity.
Employee behavior remains a weak point. Workers can upload confidential institutional documents to public AI platforms without knowing how that information may be accessed or reused. Employee education and enforcement of security protocols are therefore core controls, not administrative details.
For product and operations teams, the test is practical: contain the breach, keep the service running and recover without exposing more data.
How large is the payment risk?
David Isiavwe, ISSAN’s president, said AI lets criminals act with more speed and sophistication. Deceptive messages, cloned voices, synthetic videos and social engineering become harder to detect, while an incident at one interconnected institution can affect organizations and countries beyond the immediate target.
He called for stronger public-private partnerships, intelligence sharing and prompt action on identified threats. He said trust remains central to the digital economy and customers must be able to rely on the security and integrity of the financial and other digital services they use.
Dele Alabi, president of the Chartered Institute of Bankers of Nigeria, echoed the warning: AI improves financial-service efficiency and innovation while giving criminals synthetic identities, deepfakes and automated social engineering tools. A breach in Nigeria’s interconnected financial ecosystem could spread risks across institutions and markets.
Nigeria processed about 47.88 billion electronic payment transactions worth N3.46 quadrillion in 2025, Alabi said. Industry efforts reduced electronic-payment fraud losses by more than 50%, to N25.85 billion in 2025 from N52.26 billion in 2024, though he said unreported incidents mean the total may be understated.
The stakes have moved from an IT issue to a financial-stability issue. Alabi asked bank boards to invest in cyber resilience, management teams to put cyber risk into enterprise-wide decisions, and employees to follow security protocols. Ethical leadership, employee vigilance and cooperation among banks matter as much as technical controls.
The three speakers’ shared position was to balance innovation with security, accountability and resilience. Perfect prevention may be impossible, but as AI enters more product workflows, teams will need bounded autonomy, verifiable identity and tested recovery that protect critical services and public confidence.
via businessday.ng (Source)
More from Daniel Okafor
Show full bio
Staff writer covering technology and product news in product management at Roadmap File.
60 articles