In May 2025, days after India’s air strikes on targets inside Pakistan, a message began moving through WhatsApp groups across northern India: automated teller machines, it claimed, would be shut for two to three days. Nothing had happened to a single ATM, bank server or payment rail.¹ The government’s own clarification blamed the rumour on a broader wave of digital misinformation that it attributed to actors operating out of Pakistan, timed to follow the strikes and a run of setbacks for the Pakistani military. No system was breached and no interface was disabled. Yet for several days a meaningful share of a fully functioning banking public behaved as though the network might not be there when they needed it. That gap, between a system that works and a public no longer sure it does, is where South Asia’s hybrid warfare contest is increasingly being fought.

A companion analysis from the South Asia Council for Security Policy (SACSP) on the region’s Digital Public Infrastructure² has already argued that state and criminal actors are shifting attention away from grids and pipelines and toward Aadhaar, UPI and the identity systems built on their model. That is one level of the same contest. There is a level above it: an attacker does not need to breach a system at all if enough people can be convinced, even briefly, that it has already been breached. Three points follow. First, India’s DPI stack now works as a belief system as much as a technical one, so the country’s strategic centre of gravity is shifting toward digital trust itself. This is closer to what NATO’s own doctrine writers call the “cognitive” domain³ than to anything in a conventional cyber-defence manual.⁴ Second, hostile actors are combining ordinary cyber activity with AI-generated information operations to attack four distinct dimensions of that trust: algorithmic credibility, transactional integrity, identity assurance and public confidence. Third, the regulatory architecture built to protect the confidentiality, integrity and availability of systems has no real vocabulary for protecting perception or legitimacy. A Digital Trust Assurance framework is meant to close that gap.

From Grid to Ledger to Belief

For most of the last decade, hybrid warfare in South Asia has focused on physical assets. Typical examples are the capture of a control centre, damage to the electric power supply, or a blackout across the whole grid. Attention is now moving up to the identity and transaction layers of DPI. The 2023 CoWIN breach, in which a Telegram bot returned citizens’ Aadhaar numbers, passport details and vaccination records, is one example.⁵ It shows that interoperability, DPI’s greatest strength, is also its greatest weakness. The next step goes beyond the identity layer to the cognitive layer. A security model built on confidentiality, integrity and availability (the CIA triad) can make sure outsiders cannot read or alter information and that systems stay online. It cannot stop an adversary from convincing the public that none of these protections holds, even when all three do. NATO’s Allied Command Transformation has spent several years developing a formal concept for exactly this problem. It defines cognitive warfare as an attempt to change how people think, which weakens institutions and erodes trust in a society without any contact with its information systems.³ Indian regulators were not set up with this threat in mind, and neither was most of the literature on DPI security.

The Four Fronts of Digital Trust

Algorithmic credibility. The clearest evidence that this shift is under way came in India’s 2024 general election, the largest election in history. Observers at the time asked whether deepfakes could undermine trust in the electoral system itself.⁶ Synthetic media stopped being a curiosity and became a working campaign tool.⁷ None of this required hacking a database. It only required that viewers be unsure, for a few seconds before they scrolled on, whether a video was real. India has begun to respond. Amendments to the Intermediary Guidelines, finalised by the Ministry of Electronics and Information Technology (MeitY) in February 2026, require synthetically generated content to carry clear and prominent labels⁸ and, where feasible, embedded metadata identifying the tool used to create it. Digital rights groups have pushed back. They argue that the definition of “synthetically generated information” is vague enough to work as a general-purpose takedown power rather than a narrow transparency measure.⁹ That is a fair concern, given how broadly platforms could be asked to interpret it. Both critiques can be true at once: the absence of any labelling regime left algorithmic credibility undefended for years, and a poorly bounded one risks becoming a censorship tool rather than a trust-restoration one.

Transactional integrity. UPI’s scale is itself the vulnerability. The network carried 185.8 billion transactions in the 2024-25 financial year, a 41.7 per cent increase on the year before,¹⁰ and now accounts for more than four-fifths of India’s digital payment volume. Fraud has grown roughly in step. One industry estimate put reported UPI fraud at over 630,000 cases in a recent period, worth close to ₹4.85 billion, or roughly 0.4 per cent of transaction value.¹¹ That is a small share of an enormous base, but a large and rising absolute number. What matters strategically is less the loss itself than who bears it. A 2025 LocalCircles survey found that one in five families with a UPI user had experienced fraud at least once in the previous three years, and that about half of victims never reported it at all.¹⁰ This suggests the official figures understate the true scale by a wide margin. A payment system does not need to fail technically to lose its social licence. It only needs enough of its users to decide privately that tapping “pay” carries real risk.

Identity assurance. SACSP’s companion analysis of the region’s DPI weaponisation has already documented the identity layer’s exposure at length, from the CoWIN leak⁵ to Nepal’s under-verified National Identity Card portal.² What deserves separate attention here is a narrower and newer risk. Identity assurance increasingly depends on biometric or voice-based authentication that generative AI can now forge convincingly. Indian authorities have already flagged the emergence of illicit services offering AI-altered voice cloning for financial fraud and impersonation of public figures.¹² Applied to voice-based banking authentication or video KYC, such a capability would attack identity assurance from the inside, with no database leak at all. A leak tells an attacker what a person’s identity is. A convincing forgery lets an attacker perform that identity without ever touching the record that describes it.

Public confidence. The oldest of the four fronts is also the hardest to regulate away. India’s viral-rumour problem long predates generative AI. In November 2016, after demonetisation, one hoax claimed that the new currency notes carried tracking chips, and another rumour of an imminent salt shortage triggered stampedes outside grocery stores in four states.¹³ Both show that a population primed to distrust official channels will act on a falsehood within hours, with or without AI. What has changed is speed and deniability. The 2025 ATM rumour that opened this article followed the same mechanics as the 2016 cases, but it arrived amid an acknowledged foreign disinformation push timed to a live security crisis.¹ Public confidence is therefore not a background condition that DPI needs in order to function. It is an attack surface in its own right, one that a hostile state can time deliberately rather than merely exploit opportunistically.

Doctrine Built for the Wrong Threat

None of India’s main digital governance authorities was set up with this threat model in mind. The Indian Computer Emergency Response Team (CERT-In) works mainly through its 2022 directions, which require organisations to report incidents within six hours of noticing them.¹⁴ That is a response to events that have already happened, not continuous monitoring of an adversary’s intent. The MeitY rules on synthetic content cover labelling and disclosure.⁸ They address the credibility of algorithms but say nothing about transactional integrity, identity assurance or the timing of rumours. The rules also focus on consent and grievance redress. As a result, they treat a state-sponsored disinformation campaign in the same way as an ordinary user forwarding a fake message without checking it.

Toward a Digital Trust Assurance Framework

A workable approach treats digital trust as an asset that needs its own assurance framework, built around three groups of activity. The first is cyber resilience, which India already does reasonably well: protecting the confidentiality, integrity and availability of platforms through existing security measures. The second is information integrity. This starts at the platform layer but extends to the content layer, and includes mandatory provenance labelling of synthetic media and public fact-checking. The Misinformation Combat Alliance’s Deepfakes Analysis Unit already runs a public WhatsApp tipline.¹⁵ Information integrity also needs agreed rapid attribution protocols, so that the government can answer unfounded rumours quickly and with correct attribution, rather than improvising each time. The third is the governance of AI systems used for fraud detection and content moderation. This is a new and lightly regulated area, yet these systems now help the public trust the country’s identity and financial infrastructure.¹²

India needs one body to coordinate MeitY, CERT-In, the RBI and the Election Commission. At present, no institution looks across identity fraud, payment fraud, electoral misinformation and platform conduct together. Trust erosion should also be treated as a harm in its own right, because hostile actors deliberately work to create distrust in parts of DPI whether or not any system fails. It needs its own detection and response mechanism.

The Coming Contest for Belief

Taken alone, each rumour, deepfake or fraud looks like an isolated incident that cannot seriously threaten digital public infrastructure holding the identities, financial records and health records of over one billion people. That is where the danger lies. Each one is a small event that stays below the level that would trigger an official response, but each chips away at the belief that the infrastructure can be relied on. As instances of apparent failure multiply, they add up to a message that matters more than any single incident: the public no longer assumes that its government’s systems work.