From banking Trojans that automate transfers within authenticated sessions, using overlays and abusing accessibility services, to the Malware-as-a-Service model exemplified by DroidBot, recent developments expose the limitations of a debate focused on one-time passwords (OTPs) and user fault. Platform-level defences (Google Play Protect) and new European rules (Verification of Payee and DORA) raise the standard, but do not fully address a crucial layer: the security of the financial application itself. This article argues that client-side hardening, starting with preventing other applications from displaying overlays, forms part of the “appropriate security measures” that payment service providers must be able to demonstrate under Italian Legislative Decree No. 11/2010, with direct implications for the burden of proof and the assessment of gross negligence. The prevailing account of mobile banking fraud still centres on phishing: a text message impersonating a bank, a fake webpage, or a telephone call inducing the victim to disclose a one-time password. Under that model, legal debate focuses on the user’s conduct and the threshold for gross negligence, because the user is the person who “hands over” the credential. The figures confirm the importance of this pattern. According to the Bank of Italy’s Report on Complaints by Customers of Banks and Financial Companies (June 2026), there were 750 fraud-related complaints in 2025, an increase of 2% over the previous year. Just under half concerned the theft of login credentials, transaction authentication codes or victims’ identification data; within this category, 57% of cases involved social engineering, smishing, vishing and phishing. These figures depict a phenomenon still largely driven by deception of the user. A significant part of the threat, however, has shifted elsewhere. In more advanced banking Trojans, the OTP is not handed over: it is intercepted, while the transaction is controlled directly on the device. A typical attack vector is the abuse of Android’s AccessibilityService. Originally designed to support accessibility, this functionality enables malware to read screen content and perform taps and gestures automatically, entering and confirming a bank transfer within a session already authenticated by the user. This technique is known as an Automated Transfer System (ATS). It is often combined with overlay windows used to capture credentials and with the interception of SMS messages and notifications to steal one-time codes. A real-world sample recently examined in a forensic context illustrates this model. It presented itself using the name and icon of a widely used legitimate application, but had a randomly generated package name and a debug signature rather than a production signing certificate. The permissions list alone revealed the fraud chain: accessibility service, permission to draw overlay windows, reading, receiving and sending SMS messages coupled with a request to become the default messaging application, device administration privileges for persistence, and a command-and-control channel based on the MQTT protocol. The code contained typical indicators of accessibility-based automation, including reading on-screen nodes and programmatically generating gestures. The archive was also packaged with an anomalous flag that caused many analysis tools to refuse to read it, although it remained fully installable and executable by the operating system: an anti-analysis technique. These permissions—accessibility, SMS reading and notification access—are precisely those that platforms now regard as high-risk, because they enable the interception of OTPs and observation of screen content. A sample displaying these characteristics falls within the category of banking remote-access Trojans (RATs) distributed as Malware-as-a-Service (MaaS). A particularly relevant example is DroidBot, documented by Cleafy in December 2024: a remote-access Trojan offered by subscription for approximately USD 3,000 per month to various affiliate groups, with a builder enabling them to generate customised builds and evade detection. DroidBot combines overlays, keylogging, SMS interception and automated interactions (ATS), and has a distinctive feature: dual-channel communication, with data exfiltration via MQTT and commands received via HTTPS. This provides a useful framework, but compatibility must not be confused with attribution. Overlapping techniques alone do not establish the malware family. The MQTT channel, for instance, has also been observed in the Copybara and BRATA Trojans; attribution requires examination of configuration details (broker and topics), command channels and known signatures. Such evidential rigour distinguishes a defensible classification from a premature label. The threat is not uniform. Alongside “heavyweight” Trojans employing overlays and accessibility services, Cleafy documented an active campaign in Italy in 2025 known as SuperCard X: a Malware-as-a-Service platform that exploits NFC relay technology to enable fraudulent payments and cash withdrawals by inducing the victim to bring a payment card close to an infected phone. This represents the opposite approach: deliberately minimal malware, without high-risk permissions or overlays, designed to remain below heuristic detection thresholds. These two contrasting approaches converge on one point: some forms of fraud bypass server-side safeguards altogether. Platform providers have responded. According to Google, Play Protect’s anti-fraud protection scans in real time applications installed through sideloading sources—browsers, file managers and messaging applications—and blocks installation when they request sensitive permissions such as receiving or reading SMS messages, accessing notifications or using accessibility services, which are commonly abused to steal codes and monitor the screen. The feature has expanded from an initial pilot to 185 markets, covering more than 2.8 billion devices. Android 16 adds protections during installation and when accessibility features are activated during a telephone call, as well as an automatic prompt to stop screen sharing at the end of the call. These are significant measures, but they operate at operating-system level: their effectiveness depends on the Android version, device configuration and whether the user manually overrides an installation warning. They do not replace what a financial application can—and arguably should—do to protect its own sensitive screens. Platform safeguards reduce the attack surface; they do not eliminate it. Here the issue becomes concrete and verifiable, because standardised countermeasures exist. The OWASP Mobile Application Security Testing Guide (MASTG) sets out an explicit hierarchy of defences against overlays and tapjacking. The strongest is to declare the HIDE_OVERLAY_WINDOWS permission and invoke setHideOverlayWindows(true), available from Android 12 (API 31), which hides all non-system overlay windows while the application is in the foreground. Secondary measures, including for backwards compatibility, are filtering touches when a view is obscured (filterTouchesWhenObscured) and using FLAG_SECURE on sensitive screens. It should also be recalled that Android 12 blocks fully occluded touch attacks by default, subject to the qualification that, for system alert windows, blocking applies only above a specified opacity threshold. There is, however, a technical nuance that must be acknowledged if the analysis is to be more than a slogan: blocking overlays is necessary but not sufficient. Abuse of accessibility services does not necessarily involve drawing a window over the application; it can read the view hierarchy and inject actions, bypassing ordinary anti-overlay flags. An application handling funds should therefore also continuously monitor active accessibility services on the device, implement device binding to prevent stolen credentials from being reused elsewhere, and employ integrity attestation and runtime application self-protection mechanisms. In other words, preventing overlays is a minimum threshold, not a complete defence. At European level, Regulation (EU) 2024/886 on instant payments introduced Verification of Payee (VoP), by inserting Article 5c into Regulation (EU) No. 260/2012. Since 9 October 2025, VoP has been mandatory in the euro area for euro credit transfers, both instant and standard. The system compares the stated IBAN with the beneficiary’s name and returns an outcome (match, close match, no match or verification not possible). This is a useful safeguard, but the legislature deliberately made it informational rather than blocking: the final decision remains with the customer. More importantly, verification addresses IBAN spoofing “upstream”, when the user enters the beneficiary details. It is much less effective against an ATS attack operating within an authenticated session and manipulating the transaction on the device. This is a further argument in favour of client-side protection. As regards resilience, Regulation (EU) 2022/2554 (DORA), applicable since 17 January 2025, requires financial entities to adopt an ICT risk-management framework that emphasises security by design. Within that framework, hardening the mobile application can be understood not as an optional extra but as part of a broader obligation of operational robustness (although the precise scope of application must be assessed case by case). Under the framework established by Italian Legislative Decree No. 11/2010, where a payment transaction is disputed, the payment service provider must first prove that the transaction was authenticated, accurately recorded and entered in the accounts, and was not affected by a technical breakdown or other deficiency. The mere recording of the use of a payment instrument is not, in itself, sufficient to prove that the transaction was authorised by the payer or that the payer acted fraudulently or with gross negligence. The Italian Banking and Financial Ombudsman (Arbitro Bancario Finanziario, ABF) has further clarified that proof of strong customer authentication is a logical prerequisite to assessing gross negligence: its absence, even in part, is decisive and leads to reimbursement. Conversely, where an intermediary seeks to rely on the customer’s fraud or gross negligence, it must provide specific factual evidence from which that conduct may be inferred, including by presumption (ABF Coordinating Panel, Decision No. 22745/2019). The Italian Supreme Court of Cassation, after an earlier phase in which it treated banks’ liability as approaching strict liability, has adopted a more balanced approach, excluding intermediary liability where the customer is grossly negligent (including Cassation Decisions Nos. 7214/2023 and 3780/2024). Such negligence nevertheless requires inexcusable imprudence and may be established through serious, precise and consistent presumptions. A particularly relevant line of decisions concerning the scenarios described here follows the same reasoning: the fact that a transaction can technically be traced to the user’s device does not, by itself, make it authorised. Where the fraudster prepares the transaction and the victim merely confirms it—through a push notification, biometrics or a code—believing that they are validating a legitimate action, the free and informed consent required by Article 5 of Legislative Decree No. 11/2010 is absent, and the transaction cannot be considered authorised (see, among recent decisions, ABF Milan Panel, Decision No. 3501/2025). This principle directly affects malware-based ATS attacks, in which authentication may appear formally valid on the customer’s device without reflecting any genuine intention to make the payment. Our proposed interpretation has two elements. First, the adequacy of security measures should not be assessed solely by reference to authentication, but also in light of the application’s ability to withstand known, technically documented attack methods that can be countered through safeguards available under recognised standards. From this perspective, application hardening—including the anti-overlay and anti-accessibility-abuse countermeasures identified by OWASP as good practices—may contribute to the overall assessment of whether the provider’s security measures are appropriate. Second, the user’s conduct in an ATS attack must be assessed differently. If the OTP is intercepted rather than handed over, and the transaction is automated through an abused accessibility service, gross negligence cannot be evaluated as though the case were conventional phishing. The ABF’s case law is already developing this distinction: in social-engineering fraud cases, it tends to exclude gross negligence where there are no concrete warning signs perceptible to the user. That approach places an even greater evidential burden on the intermediary in malware-based ATS scenarios, where nothing is “handed over” and no anomaly is visible (see, among recent decisions, ABF Rome Panel, Decisions Nos. 8658/2025 and 10894/2025). Mobile payment security is a multi-layered system involving the user, the platform and the bank. Focusing the assessment solely on the human element, or relying exclusively on the operating system, leaves exposed the very layer that controls the funds. The right question—technical before it is legal—is not only “what should the user have done?”, but also “what baseline level of security can reasonably be expected from an application that transfers money?”. Answering that question through verifiable standards rather than generic formulas is now an integral part of assessing whether security measures are appropriate. • Bank of Italy, Report on Complaints by Customers of Banks and Financial Companies (June 2026) — 2025 data on fraud complaints and credential-theft techniques; also reported by Il Sole 24 Ore, 25 June 2026 • Google, Keeping Google Play & Android app ecosystems safe in 2025 — security.googleblog.com / blog.google • Google, What’s new in Android security and privacy in 2025 — blog.google (Android 16 protections concerning sideloading, accessibility during calls and screen sharing) • OWASP Mobile Application Security, MASTG-BEST-0040: Preventing Overlay Attacks and MASTG-KNOW-0022 — mas.owasp.org • Android Developers, Tapjacking (security/risks) — developer.android.com • Cleafy Labs, DroidBot: insights from a new Turkish MaaS fraud operation (December 2024) — cleafy.com • Cleafy Labs, SuperCard X: exposing a Chinese-speaker MaaS for NFC Relay fraud operation (2025) — cleafy.com • Regulation (EU) 2024/886 (instant payments; Article 5c of Regulation (EU) No. 260/2012); Bank of Italy, information on Verification of Payee (VoP) • Regulation (EU) 2022/2554 (DORA) • Italian Legislative Decree No. 11 of 27 January 2010 (implementing PSD2); Directive (EU) 2015/2366 (PSD2) — Articles 10–12 on the burden of proof and liability for unauthorised transactions • Italian Banking and Financial Ombudsman (ABF) — Coordinating Panel Decision No. 22745/2019 (burden of proof/gross negligence); Rome Panel Decisions Nos. 8658/2025 and 10894/2025 (social-engineering fraud) — arbitrobancariofinanziario.it database • Italian Supreme Court of Cassation, Order No. 7214/2023 and Judgment No. 3780/2024 (exclusion of bank liability in cases of customer gross negligence); earlier decisions Nos. 10638/2016, 2950/2017 and 9158/2018 • Bank of Italy, Report on Fraudulent Payment Transactions in Italy in the Second Half of 2024 (August 2025) • G. (Diritto Bancario), Le frodi informatiche negli ultimi orientamenti dell’ABF — dirittobancario.it (review of decisions; references to individual decisions to be checked)From Smishing to On-Device Automation
What the Operating System Already Does (and Why It Is Not Enough)
What the Financial Application Should Do
The Regulatory Framework: Verification of Payee and Operational Resilience
The Legal Issue: “Appropriate Security Measures” and the Burden of Proof
Shared Responsibility
Sources