Something close to a consensus is forming among Western intelligence establishments: artificial intelligence is the future of espionage. From Langley to London, from Brussels to Canberra, senior officials and policy strategists have spent the better part of this decade making the case that AI will revolutionise how states collect, process, and act on intelligence. The Office of the Director of National Intelligence in the United States has made AI integration a centrepiece of its modernisation agenda. The UK’s GCHQ (Government Communications Headquarters) has invested heavily in machine learning for signals analysis. The European Union’s intelligence apparatus increasingly relies on algorithmic tools for open-source surveillance. Even China’s Ministry of State Security (MSS) has fused mass cyber-espionage with AI-driven data processing into one of the most formidable intelligence machines in modern history. The momentum behind this global shift is real, and it is accelerating.
But momentum is not the same as wisdom. And a growing body of serious scholarship is issuing a warning that deserves careful attention from Dhaka: the wholesale embrace of AI in intelligence operations carries risks that are not merely technical but strategic, and in some cases fatal. For Bangladesh, a country whose intelligence agencies were built on human networks, whose regional security environment is uniquely complex, and whose institutional capacity for AI governance remains nascent, following the global trend without scrutiny could prove to be a catastrophic error.
The most rigorous recent articulation of this warning comes from Thomas Mulligan, a researcher at the RAND Corporation and a former CIA case officer, writing in Studies in Intelligence, the CIA’s flagship peer-reviewed journal, in March 2026. Mulligan’s argument is counterintuitive but carefully reasoned. He does not deny that AI will transform intelligence work. His claim is more consequential: that precisely because AI is transforming technical collection by making satellite imagery, signals interception, and open-source analysis cheap and accessible to any competent actor to the relative value of human intelligence, HUMINT, is rising, not falling. When every state and non-state actor can deploy comparable technical tools, the marginal advantage belongs to whoever can still run a trusted human source inside a hostile organisation. Bangladesh’s neighbours and adversaries understand this. Dhaka should too.
Mulligan’s second argument is perhaps more immediately alarming for a country surrounded by sophisticated regional actors. AI has become an extraordinarily effective engine of disinformation. Large language models (LLMs) can generate false but internally consistent intelligence reports. Generative AI can produce deepfake audio, video, and documents that are, to the untrained eye and increasingly to the trained one, indistinguishable from genuine material. Mulligan describes a scenario that should unsettle any intelligence planner: an adversary that floods the signals environment with ten times the volume of genuine traffic, each item a convincing deepfake, rendering electronic collection not merely unreliable but actively counterproductive. In such an environment, a human source physically present within the adversary’s structures becomes the only reliable instrument of verification. No algorithm can replicate that.
This is not a distant hypothetical for South Asia. Deepfake content globally surged from approximately 500,000 files in 2023 to a projected eight million in 2025, with the Asia-Pacific region recording a 1,530 percent rise in deepfake incidents over comparable timeframes, according to industry data. In July 2024, a deepfake audio impersonating Philippine President Ferdinand Marcos Jr, closely mimicking his voice and purportedly ordering a military strike against China, circulated widely, causing genuine public panic. The Global South Forum has documented the systematic use of such tools in geopolitical disputes across the region. Bangladesh, which shares extended borders with two major powers like India and China, also sits at the edge of the unresolved Myanmar crisis, is an obvious target for exactly this kind of operation. An intelligence community that has pivoted toward AI-mediated collection, without the human networks to corroborate what it receives electronically, would be dangerously exposed.
The scholar Matt Chessen, writing for the Atlantic Council, has made a related argument that bears directly on Bangladesh’s predicament. Chessen observes that AI tools, when provided with rich data about individuals, can decode human psychology with a precision that exceeds self-knowledge of identifying personalities, fears, and motivations with a granularity no human analyst can match at scale. This capacity for micro-targeting, Mulligan notes, has a paradoxical epistemic consequence: as AI-powered influence becomes more pervasive and more effective, the rational response is to treat all electronically mediated communication as lower-trust by default. The logical endpoint, which researchers and futurists have described as “epistemic collapse”, it is a world in which digital communications are presumed unreliable unless corroborated by face-to-face human contact. The ancient tools of espionage, dead drops, brush passes, and clandestine meetings are not relics in this world. They are the last line of epistemic defence.
The limitations of AI in intelligence are not only operational; they are conceptual. Zachery Tyson Brown, writing in Studies in Intelligence in 2024, argued that no matter how large an AI model becomes, it will never encompass the world in its totality, where there is no quantity of data that permits the reliable forecasting of genuinely novel events in a complex competitive environment. Brown’s argument is a philosophical corrective to the techno-optimism that has captured so many security establishments. Intelligence, at its most consequential, is about precisely those novel events: a sudden shift in leadership intentions, a secret decision made in a closed room, a covert agreement between two hostile powers. These are things that AI can only ever infer from patterns. A human source can know them directly.
For Bangladesh’s intelligence agencies like the Directorate General of Forces Intelligence (DGFI), the National Security Intelligence (NSI), and the Special Branch, the stakes of this debate are acute. Recent reporting indicates that deepening defence ties with China, including trilateral security frameworks established in June 2025, are accelerating the acquisition of signals intelligence tools and transitioning agencies away from their traditional HUMINT dominance. If this transition is managed without deliberate institutional investment in human intelligence capabilities, Bangladesh risks arriving at exactly the vulnerability Mulligan describes: an intelligence community technically capable of processing large volumes of electronic data, but unable to verify which of it is real, which is fabricated, and which has been crafted specifically to mislead it. The result would not be a stronger intelligence apparatus. It would be a more expensive and more confident one that is equally blind.
The experience of Bangladesh’s regional neighbours is instructive and sobering. A 2026 analysis in Global Security Review observed that India’s Research and Analysis Wing (RAW) and Pakistan’s Inter-Services Intelligence (ISI), the two agencies that have most shaped the South Asian intelligence environment, have built their formidable reputations on deep human networks cultivated over decades. Their current challenge is not whether to replace HUMINT with AI, but how to integrate fragmented AI capabilities into legacy human structures without destroying the institutional knowledge that makes those structures valuable. Bangladesh cannot afford to skip this foundational stage. The agencies that command regional influence did not achieve it through algorithms. They achieved it through officers who knew their sources personally, who had earned trust over years of careful relationship-building, and who could read a meeting room in ways no machine has yet learned to replicate.
None of this is an argument that Bangladesh’s agencies should ignore AI altogether. Mulligan is explicit that AI has genuine and valuable applications in intelligence work: screening problematic walk-in sources, planning surveillance detection routes, processing open-source data, and assisting in agent validation. The Special Competitive Studies Project’s 2025 white paper on the “Digital Case Officer” frames the future of espionage explicitly as a human-machine team, not a machine replacement. The distinction matters enormously. AI as an augmentation of skilled human officers is a force multiplier. AI as a substitute for them is a strategic liability, particularly in an environment where adversaries are specifically designing their operations to exploit the vulnerabilities of AI-dependent systems.
The global intelligence community’s enthusiasm for artificial intelligence is understandable, and in some contexts justified. But Bangladesh is not the CIA. It is not the MSS. It does not have the institutional depth, the training pipelines, the AI governance frameworks, or the counterintelligence culture to safely absorb a rapid pivot away from human intelligence. What it does have, built over fifty years, at considerable cost and effort, is a set of human networks and operational traditions that represent a genuine national security asset. To dismantle or neglect that asset in pursuit of algorithmic efficiency, at the precise moment when AI is making human verification more valuable than ever, would be an act of strategic self-harm. The machines are getting smarter. That is exactly why Bangladesh still needs its human spies.
Arman Ahmed
Arman Ahmed is an independent researcher and a research analyst at the Nicholas Spykman International Center for Geopolitical Analysis in Paris, where his work focuses on South Asian geopolitics and international security. His writing has appeared in The Geopolitical Monitor, The Geopolitics, Modern Diplomacy, The Madras Gazette, Eurasia Review, The Daily Star, The Business Standard, Prothom Alo English, the Australian Institute of International Affairs, and other national and international outlets. He specializes in strategic affairs, South Asia, defence policy, and foreign policy analysis.
