National CERT warns banks that a false balance can cause panic, and Pakistan's apps carry 2.9 billion payments a quarter

On 5 November Pakistan's National Cyber Emergency Response Team, known as National CERT, warned banks about an attack that changes the balance a customer sees while the money stays where it is. Its advisory, numbered NCA 26.051126, lists manipulation of displayed account balances among the risks for financial institutions and says it "may trigger customer's panic even where underlying funds remain same."

The advisory is built around artificial intelligence. It says growing use of AI, automated bots, botnets and agentic tools to automate reconnaissance, find vulnerabilities and run credential attacks may let attackers widen the scale, speed and sophistication of their operations. It applies to all sectors but singles out two: financial services, and operational technology and critical infrastructure.

For banks it names core banking systems, digital banking platforms, APIs, payment systems and customer facing applications as targets. Its warning signs are automated login attempts, repeated authentication failures and credential stuffing, unusual API requests, abnormal transaction patterns, unauthorised access to payment systems, and any difference between core banking records and the balances customers see.

A forecast, not an incident

The advisory does not report an AI driven attack in Pakistan. Its only example is a set of recent attacks on refrigeration and freezer systems at European supermarkets, and it says AI involvement in them has not been confirmed. It is a warning about capability, and it asks organisations to check their preparedness against bot driven attacks.

The defences it asks for are conventional: multi factor authentication, protection against automated logins, privileged access management, monitoring of bot activity and API traffic, integrity checks between customer facing applications and core banking records, and independent transaction reconciliation. For operators of utilities and industrial plant it adds limiting internet exposure, segmenting operational technology from office networks, and keeping backup operating procedures.

“A compromise could disrupt physical processes and critical services.”

Why a wrong balance matters

A banking app does not hold money; it asks a server what the ledger says and draws the answer. An attacker who alters that answer in the display layer, such as an interface gateway, a cache or a compromised mobile backend, leaves the ledger correct and moves no funds. The harm is the reaction of customers who see a wrong figure, which is what the advisory's reference to panic describes. The integrity check it recommends compares what the app shows with what the core banking record holds.

The State Bank's payment figures for January to March 2026 show the scale of the display layer. Banks, wallets and electronic money institutions handled 3.7 billion retail payments worth Rs168.8 trillion in the quarter. Mobile apps carried 2.9 billion of them, worth Rs42 trillion. Over the 90 days that is about 32 million app payments a day, or roughly 370 every second on average. The average app payment was about Rs14,500.

Compare the channels customers could turn to. The State Bank counts 20,232 bank branches and 819,397 banking agents. Branches processed 128 million transactions worth Rs99.5 trillion, an average of about Rs777,000 each, and agents 155 million worth Rs1.1 trillion, about Rs7,100 each. Together they handled 283 million transactions, about 3.1 million a day. As arithmetic, not a forecast: if 10 percent of one day's app payments, about 3.2 million, turned up as branch or agent visits, that network would see roughly double its normal daily load. Branches also hold most of the value, Rs99.5 trillion or about 59 percent of all retail payments by value, against about 25 percent for apps.

Old attacks, faster

In practice the AI label describes speed and volume, not a new class of attack. Credential stuffing, where bots replay stolen passwords across thousands of logins, and automated probing of interfaces have existed for years. Automation lets an attacker try more combinations and adapt faster, which is why the standard defences matter most at the login and the interface: rate limits, bot detection and multi factor checks.

Institutions that already reconcile their display layer against the ledger in real time gain most from the advisory. The weaker position belongs to those that buy their app backend from outside suppliers, a category the advisory flags when it asks banks to check third party integrations. The advisory is guidance, not a regulation, and sets no deadline. Its one concrete request to every organisation is to keep security logs and report suspected incidents through the national cyber security channel.