When Love Becomes a Language Model: Inside the Industrial Romance Scam

May 19, 2026

In February 2026, Erin West walked through an empty scam compound in Sihanoukville. The lights, she would later tell the Stolen podcast audience, were off across most of the Cambodian coastal city that had, for the better part of a decade, served as the operational headquarters of the global romance-fraud industry. Walking the corridors of a half-abandoned dormitory building, she found the small evidence of recent occupation: bedding still bunched on the floor, foreign-language posters peeled from concrete walls, plastic wash basins stacked in the corners. She also found, scattered around the workstations, the script binders and conversation guides that the trafficked workers had been required to memorise. What she did not find, by and large, was the workers themselves. They had been moved.

West, a former Santa Clara County deputy district attorney who spent twenty-six years prosecuting tech crime before founding Operation Shamrock, has spent three years documenting an industry that, on any honest accounting, has surpassed in scale and harm the international drug trade. Pig-butchering operations alone, the long-con romance fraud that combines synthetic intimacy with sham crypto investment, generated an estimated forty-three point eight billion dollars in revenue across Burma, Cambodia and Laos in 2023, equivalent to roughly forty per cent of those countries' combined official gross domestic product, according to figures cited in testimony to the United States Congress. The industry has not been dismantled by the international crackdown of late 2025. It has migrated, restructured and, increasingly, automated.

That last word is the one this story turns on. The compounds West walked through were emptier than she expected because the work being done inside them is, slowly but visibly, being absorbed by software. The cost of buying a trafficked Vietnamese teenager, beating compliance into him with rebar, and chaining him to a workstation to run twenty-four parallel romance scripts across WhatsApp and Tinder turns out to be higher than the cost of a stack of GPUs running an open-weights large language model fine-tuned for emotional manipulation. The fraud has not gone away. It has, in the way of all industrial production, become more efficient.

And the bills have started arriving.

The Numbers That Made It a Public Crisis

In February 2026, an industry analysis circulated by Credit Union Today put the headline figure for American losses to AI-driven romance fraud at six hundred and seventy-two million dollars across the preceding year. The figure tracked closely with the FBI's own 2025 Internet Crime Complaint Center reporting, which recorded romance and confidence scams as among the highest-loss categories of cybercrime, and which for the first time in IC3's twenty-five-year history saw total internet-crime complaints crossing one million. Reported fraud losses across all categories climbed to twenty point nine billion dollars, a twenty-six per cent increase over 2024.

The British numbers tell a similar story at a smaller scale. UK Finance, the industry body that tracks payment-system fraud across British banks, reported that romance scams cost UK victims twenty point five million pounds in the first six months of 2025 alone, across nearly three thousand cases, a thirty-five per cent increase year on year. City of London Police, which leads the national police response on fraud, recorded more than one hundred and six million pounds lost to romance fraud over the 2024 to 2025 financial year. TSB Bank, in February 2026, warned that romance fraud was rising sharply across its customer base, with the average victim sending eleven separate payments, losing roughly seven and a half thousand pounds before the deception was discovered.

These numbers are, in every estimation produced by every researcher who has examined them, undercounts. AARP's Foresight 50+ Omnibus survey, conducted by NORC at the University of Chicago and published in February 2026, found that more than half of American adults who had lost money to a scam never reported it to anyone. Among those who did report, only twenty-six per cent went to law enforcement; only twenty-three per cent contacted their bank. The reason, AARP's research consistently finds, is shame. While sixty-two per cent of people view scam victims primarily as targets of a crime, sixty per cent simultaneously assume the victim was naïve or too trusting. Those two attitudes coexist in the same survey, often in the same respondent, and they are the social mechanism through which under-reporting becomes structural.

Professor Monica Whitty, head of department for software systems and cybersecurity at Monash University and the leading academic authority on romance-scam victimology, has been documenting this dynamic since the early 2010s. Her foundational research, conducted with Tom Buchanan at the University of Westminster, established what subsequent studies have only deepened: that for many romance-scam victims, the loss of the relationship is more upsetting than the financial loss, with some victims describing the experience as equivalent to bereavement. The double trauma of grief and humiliation creates a powerful reporting suppressor. The figures we have, in other words, are a fraction of the harm being done.

What Changed When the Models Arrived

To understand why the AI inflection of this fraud matters, it helps to understand what it replaced. Romance scams in their pre-2023 form were a labour-intensive business. A scammer, working from a script, could maintain perhaps twenty active relationships at once, each requiring hours of attention, each calibrated to the specific emotional vulnerabilities the victim had revealed in earlier conversations, each constrained by the operator's stamina, language ability, time-zone alignment and capacity to remember which victim had told them what. The bottleneck was human cognition. The product was synthetic intimacy, and synthetic intimacy was, like every artisanal good, expensive to produce.

The bottleneck has now been removed. A study published in the Asian Journal of Criminology in 2025, examining how organised criminal groups in Myanmar were deploying machine learning across their operations, found that large language models were already being used for open-source intelligence gathering and victim-network construction, achieving estimated time and cost reductions of ninety-six point six per cent and ninety-nine point five per cent respectively compared with manual research methods. AI-enabled scams in the same study were found to be four and a half times more profitable than their manual counterparts. The authors anticipated that LLMs would, in the medium term, replace human operators in the conversational stages of pig butchering as well.

That medium term has arrived. Experian's 2026 fraud-trends report, cited approvingly by the American Bankers Association in January, named “machine-to-machine” romance scams, in which fully autonomous bots maintain hundreds of simultaneous personalised relationships, as one of the top five fraud trends to watch. The bots, the report noted, “respond convincingly, build trust over time, and manipulate victims with precision and emotion.” Norton's 2026 dating-scam research found that nearly half of US online daters reported having been targeted by a dating scam, with seventy-four per cent of those targeted falling for at least part of it. Fox Business reported in 2026 that Tinder had begun mandating facial verification and was working with World, the iris-scanning identity protocol formerly known as Worldcoin, to prove that users were human at the point of sign-up. Trustpilot user reviews now estimate that up to eighty per cent of profiles on some major dating sites are fake or AI-generated, up from ten to fifteen per cent before the generative-AI boom.

The technical capabilities being deployed are not science fiction. They are commodity. An off-the-shelf foundation model, fine-tuned on a few thousand transcripts of successful and failed romance-scam conversations, will reliably produce text that is grammatically perfect in dozens of languages, emotionally coherent across long conversational arcs, capable of remembering the victim's daughter's birthday and the name of her late husband and the specific complaint she made about her sister three weeks ago. Combine that text generation with voice cloning, with face-swapping live-video, with the ability to generate consistent images of the same fictitious “soldier deployed in Yemen” or “oil engineer working off the coast of Angola” in any pose holding any object, and the verification heuristics human beings have evolved over decades of online dating collapse simultaneously. The “send me a selfie holding today's newspaper” test, as one industry commentator told the security press in 2026, is dead.

The Architecture of Manufactured Attachment

The reason romance fraud is so devastating, and the reason its industrial scaling is so alarming, is that it does not exploit gullibility in the casual sense the public uses the word. It exploits the human attachment system, which is to say the same neural and behavioural machinery that makes long-term love possible in the first place. Whitty's persuasive-techniques model, derived from interviews with victims and from analysis of hundreds of scam conversations, identifies a sequence of stages. The scammer establishes presence and idealisation. The scammer engineers reciprocal self-disclosure. The scammer creates a private emotional world to which only they and the victim have access. The scammer, by this point typically referred to by the victim as a partner, then introduces a crisis whose resolution requires money.

What an LLM brings to this sequence is not a new technique but radically improved consistency at every stage. Human scammers, working long shifts in a Sihanoukville compound, forget details. They confuse one victim with another. They lapse out of character when tired. They misjudge tone. The model does not. It maintains a coherent persona across thousands of conversational turns, recalls every personal detail the victim has volunteered, mirrors the victim's emotional register with sub-clinical precision, and never, ever loses patience. Researchers at the University of New South Wales reported in February 2026 that participants in experimental studies of AI-generated romance content struggled to distinguish it from authentic human writing, even when explicitly told some samples were synthetic.

The harm profile that follows from this is, predictably, severe. A 2025 qualitative study published in the journal Computers in Human Behavior Reports, examining the emotional, physiological, financial and legal consequences of online romance scams in the United States, found four major themes of harm: mental-health consequences including suicidal ideation, physiological health consequences including stress-related conditions, financial consequences ranging from depleted retirement savings to bankruptcy, and legal consequences including, in some cases, prosecution of victims who had unwittingly become money mules in the laundering of other victims' funds. A separate 2025 paper in the journal Victims and Offenders, which interviewed victim-survivors and their family members about the lived experience of cyber-scam victimisation, recorded what one participant called “falling into a black hole,” a phrase that became the paper's title.

The interaction of grief and shame is the part most laypeople find hardest to model accurately. Victims are mourning the end of what felt to them like a real relationship, often the most emotionally intimate of their adult lives, and they are simultaneously being asked by the surrounding culture to feel embarrassed about the relationship having existed at all. NBC News, in its earlier reporting on victim self-harm, and the AARP's ongoing Perfect Scam podcast, which has documented multiple cases of suicide following romance fraud, have together built a body of journalism that establishes the connection beyond reasonable doubt: romance investment scams combine the two leading proximate causes of suicide identified in the public-health literature, namely the dissolution of an intimate relationship and the threat of imminent financial ruin. In a substantial minority of cases, both arrive in the same week.

The Liability Vacuum

If a foreign criminal network has stolen six hundred and seventy-two million dollars from American citizens by exploiting the infrastructure of American technology firms and American payment networks, the natural question is who, if anyone, is supposed to bear legal responsibility for what happened. The answer in 2026, as a matter of law in both the United States and the United Kingdom, is approximately nobody, distributed across a chain of approximately nobodies, each of whom can plausibly point to the next link as the appropriate defendant.

In the American framework, dating-app and social-media platforms remain heavily protected by Section 230 of the Communications Decency Act, which immunises interactive computer services from liability for content posted by their users. In February 2025, the United States Court of Appeals for the Ninth Circuit, in an opinion that the Electronic Frontier Foundation argued was correctly decided, held that Grindr could not be held responsible under product-liability or negligence theories for matching a fifteen-year-old who had falsified his age with adult users who subsequently raped him. The Supreme Court declined to take up an appeal of a separate Grindr Section 230 case in 2024. The doctrinal direction of travel, despite years of political pressure from both major American parties, is that the platforms remain shielded from liability for the third-party content they host, including content generated by foreign scam operations.

Foundation-model providers occupy a different but comparable redoubt. The major American AI companies, OpenAI and Anthropic prominent among them, have invested heavily in safety training designed to refuse explicit requests to generate scam content. OpenAI's December 2025 Model Spec explicitly prohibits using the company's models for “targeted or scaled exclusion” or “manipulation” of human autonomy, and the joint OpenAI-Anthropic safety evaluation published in 2025 demonstrated that both providers' models are reasonably resistant to direct jailbreaks. The problem is that romance fraud does not require the model to produce overtly malicious content. It requires the model to produce loving content, persistently, in response to victim messages that do not in themselves trigger any safety classifier. The fine-tuning required to weaponise a model for fraud is generally trivial. Open-weights models, which can be modified offline and deployed without provider oversight, foreclose the safety-training argument entirely.

The dating platforms themselves operate in a marketing environment that requires them to insist they are doing everything possible against scammers, while a business environment that rewards reduced friction at sign-up. Tinder's introduction of facial verification and World iris-scan integration in 2025 represented the most aggressive position in the industry, but Bumble, Hinge and the rest of the Match Group portfolio have moved in the same direction more slowly. None of these platforms is currently obliged, under American law, to reimburse victims of fraud that began on their service.

The British position is, on paper, modestly stronger. The Online Safety Act 2023, fully in force across 2025 and into 2026, designates fraud as a “relevant offence” for which platforms must take proactive risk-reduction measures, with Ofcom empowered to fine non-compliant providers up to eighteen million pounds or ten per cent of qualifying global revenue, whichever is higher. Ofcom's enforcement guidance, published in late 2025, explicitly identifies romance scams as in scope. The legislation does not, however, create a private right of action for individual victims to sue the platforms directly, and as of April 2026 the regulator had not initiated headline enforcement against any major dating service for romance-scam failures.

Where the British regime materially diverges from the American one is in the payment rails. From October 2024, the UK Payment Systems Regulator's mandatory reimbursement scheme for authorised push-payment fraud requires all in-scope payment-service providers to reimburse victims of APP fraud up to eighty-five thousand pounds per claim, with the cost shared fifty-fifty between the sending and receiving banks. The PSR's first-year data, published in early 2026, indicated that eighty-eight per cent of money lost to APP scams in the first twelve months of the scheme had been reimbursed, with eighty-two per cent of claims closed within five business days. Vulnerable customers are exempt from the standard one-hundred-pound excess. For all the scheme's limitations, including a fifty-thousand-pound cap initially proposed and then raised to eighty-five thousand after lobbying from consumer groups, it represents the most substantial liability shift any major economy has imposed on the financial sector for the cost of fraud committed against ordinary citizens.

The United States has nothing comparable. American banks, under Regulation E and the broader patchwork of consumer-protection law, are generally not required to reimburse customers who authorised the transfer themselves, even if they were tricked into doing so. The cryptocurrency leg of pig-butchering fraud, which is the most common modality, is even less regulated. Tether, the issuer of the USDT stablecoin that is the rail of choice for Southeast Asian scam compounds, has frozen more than three hundred and forty-four million dollars in suspect funds in cooperation with the US Treasury's Office of Foreign Assets Control and other law-enforcement bodies, but the freezes are discretionary and post-hoc, and the funds, once moved through enough cycles, are practically irrecoverable.

The criminal-law track has, to be clear, not been entirely toothless. In October 2025, the United States Treasury and His Majesty's Treasury jointly imposed what officials described as the largest coordinated sanctions package ever directed at Southeast Asian cyber-scam networks, designating one hundred and forty-six individuals and entities tied to Cambodia's Prince Holding Group as a transnational criminal organisation. Federal prosecutors in the Eastern District of New York unsealed a twenty-six-page indictment against the group's chairman, Chen Zhi, and a separate court filing revealed the seizure of fifteen billion dollars in cryptocurrency identified as proceeds of crime. South Korea, Singapore and other jurisdictions added their own designations through November and December. Chen was arrested in Cambodia and extradited to China in early January 2026. The international press, briefly, described the action as a turning point. Erin West, a few weeks later, was watching new compounds being constructed deeper inside Cambodia, staffed by the same trafficked workers who had been moved out of the sanctioned facilities. The criminal-law approach is necessary. It is not, on the available evidence, sufficient.

What Proportionate Response Actually Looks Like

The temptation, when an industry has scaled at the speed and scope of AI-enabled romance fraud, is to reach for a single-bullet solution: ban the model, regulate the platform, prosecute the kingpin. None of these has, on its own, the dimensions of the problem. The proportionate response, if such a thing is constructible, has to operate at every layer of the stack at once, on the principle that any one defensive measure can be circumvented but enough of them in combination create friction sufficient to make the unit economics of fraud worse than the unit economics of legitimate business.

The first layer is liability allocation. Romance fraud is currently a negative externality of three industries: dating platforms, AI model providers and payment processors. None of those industries pays the cost. The victims pay the cost. Standard externality economics tells us what to do in this situation: tax the externality back onto the producers, either through direct civil liability or through mandatory insurance schemes. The British APP-reimbursement model, applied to the AI inflection of fraud, is a credible starting point. The American conversation, dominated by Section 230 absolutism on one side and reactive criminalisation on the other, has not yet caught up with the proposition that platforms hosting industrial-scale fraud should pay a proportional share of the bill.

The second layer is technical. Provenance and content-credential standards, including the C2PA work led by Adobe, Microsoft and the BBC, offer a partial defence against synthetic media in the dating context, though they require widespread adoption to matter. Real-time scam-pattern detection, of the sort already deployed by some major banks against APP fraud, can be extended to dating platforms with the cooperation of the major providers. Crucially, the friction-budget conversation needs to be reopened: dating apps that allow free-tier sign-up with minimal verification have made an explicit choice to externalise risk onto the user base, and that choice can be unwound. Tinder's iris-scan experiment is awkward and intrusive and probably necessary. The privacy trade-offs are real and demand serious public scrutiny. They do not, on the current evidence, weigh more heavily than the harm being done.

The third layer is the financial chokepoint. Stablecoin issuers, cryptocurrency exchanges and remittance corridors that process pig-butchering proceeds are the single most leveraged point in the system, because the criminal enterprise cannot survive without them. Tether's discretionary freezes are an admission that the issuer can act when it wants to. The policy question is whether we want a global stablecoin network whose anti-fraud actions depend on the goodwill of a private issuer in the British Virgin Islands, or whether we want enforceable standards. The same question applies, with greater obvious leverage, to the major centralised exchanges. Coinbase, Kraken and Binance know which wallet clusters are receiving pig-butchering proceeds. The friction-imposition tools are already built. They are not yet mandatory.

The fourth layer is shame reduction. This is the layer journalists tend to skip, because it is the least technical and the least amenable to the policy-speak the trade prefers, and it is also the layer on which everything else turns. As long as romance-scam victims feel too humiliated to report, the entire enforcement apparatus is operating on a fraction of the available signal. AARP's research finding that sixty per cent of survey respondents simultaneously regard victims as crime targets and as personally naïve is the public-attitudes equivalent of a software bug, and like a software bug it is fixable. Public-information campaigns that explicitly de-stigmatise victimhood, similar in their emphasis to the long-running campaigns around domestic violence and sexual assault, are the work of years rather than months, but they pay back at compound interest. Police forces that train officers to treat romance-scam reports as serious crime rather than personal embarrassment do better at recovery. Banks that deploy specialist fraud teams trained in conversational intervention, of the sort TSB has begun piloting in the UK, recover more money and rupture fewer relationships in the process.

The Position

The position this article arrives at is simple enough to state and difficult enough to enact. If artificial intelligence has industrialised what was once a manual fraud, then the regulatory and civil-liability response must be industrial in proportion. The decade since Section 230 was last seriously contested in Congress is the decade across which the model providers, the platform owners and the payment processors have built a fraud-vulnerable infrastructure whose marginal costs are borne entirely by people who did not consent to the trade-off. That arrangement is not, on any defensible reading of consumer-protection law or basic distributive ethics, sustainable.

The British APP-reimbursement scheme is the most concrete signal in the global regulatory landscape that another arrangement is possible. It is imperfect. It has been gamed at the margins. It has shifted some risk from victims to banks, where lobbyists are already pushing back. But it has also, in its first year of operation, returned eighty-eight pence on the pound of stolen money to ordinary people who would otherwise have lost it. That is the kind of measurable, replicable outcome that should be the baseline expectation of any modern fraud-response regime. The American conversation, which currently treats the platforms and model providers as effectively immune from civil liability for the harms that flow through their services, has not yet caught up.

What the AI-fuelled romance fraud crisis tells us, more clearly than any other AI-policy story of the last five years, is that the question of who pays is not separable from the question of who builds. The companies whose models can sustain hundreds of simultaneous synthetic intimacies have built a capability that, in the absence of corresponding obligations, will be used by whoever has the lowest scruples and the cheapest GPUs. The dating platforms whose business model depends on frictionless sign-up have built an attack surface that, in the absence of mandatory verification, will continue to be exploited by foreign criminal syndicates. The payment processors whose systems clear billions in cross-border transfers within seconds have built an exfiltration channel that, in the absence of mandatory holds and reimbursement, will continue to be the last leg of every successful scam.

There are real trade-offs in every direction. Mandatory verification at dating-app sign-up imposes real privacy costs and may foreclose use by people with good reasons to remain pseudonymous. Liability shifts to model providers risk slowing the deployment of useful AI capabilities into other domains. Bank-level holds on cross-border transfers will frustrate legitimate users with foreign relationships. None of these trade-offs is trivial. None of them is anywhere near as severe as the harm currently being externalised onto romance-scam victims, whose lives are being ended, financially and sometimes literally, in their tens of thousands every year.

Back to Sihanoukville

Walking through the abandoned compound in February 2026, Erin West observed something that has stayed with the people who watched her video diaries on the Operation Shamrock site. The compound was not closed in any meaningful sense. The workers had been moved further inland, away from the coastal compounds that had drawn the attention of Western investigators, into newer facilities in the Cambodian interior, where the construction was active and the access roads were difficult and the international press had not yet arrived. The infrastructure of synthetic intimacy was, like every infrastructure that has ever been targeted by international enforcement, relocating.

The point of the relocation, from the operators' perspective, is that the work has become much easier to relocate. A workforce of trafficked humans is heavy, slow, expensive to move and dangerous to move because moved humans speak to journalists. A workforce of language models is none of those things. It can be replicated across server farms in jurisdictions with weak enforcement, fed translated scripts in any of fifty languages, and pointed at victims selected from leaked dating-site databases or social-media data brokers. The compound in Sihanoukville with its bunched bedding and abandoned wash basins is, in retrospect, an artefact of the human-labour era of romance fraud. The coming era will leave fewer artefacts, because there will be fewer humans to leave them.

What that era requires of the rest of us is the construction of a regulatory and legal infrastructure that does not depend on the existence of a physically locatable compound full of trafficking victims. The compounds were politically galvanising, but they were also a crutch. They allowed Western policymakers to treat romance fraud as an offshore problem that would yield to offshore enforcement. The model-driven version of the same fraud will yield only to onshore measures: civil liability for the platforms and the payment rails, mandatory provenance and verification standards for the dating services, reimbursement schemes for the victims, public-information campaigns that strip the shame off reporting. None of those measures is novel. All of them are tractable. The reason they have not been built yet is not technical and not even economic. It is that the people whose retirement savings are being extracted, conversation by tender conversation, by language models running in unmarked data centres, do not yet have lobbyists who can match the influence of the companies whose infrastructure is being used against them.

That, in the end, is the actual story of the six hundred and seventy-two million dollars and the twenty point five million pounds and the unknowably larger sums that will never be reported. It is not a story about AI alignment, or about jailbreaks, or even about the trafficking compounds, which are themselves only the most visible symptom of the deeper architecture of harm. It is a story about who pays, and the answer the current system gives is the wrong one. The bills are arriving in the post, in the bank statements, in the unanswered phone calls from a partner who does not exist, and the response so far, from the institutions whose infrastructure made the fraud possible, has been a press release. It is not enough. It will not be enough until the cost of building these systems carelessly is finally borne, in cash and in court, by the people who built them.

The compound lights are off in Sihanoukville. The models are warming up everywhere else.

References

CU Today. “Love, Lies And $672 Million Lost: Inside The AI-Fueled Romance Scam Explosion.” CU Today, February 2026. https://www.cutoday.info/THE-feature/Love-Lies-And-672-Million-Lost-Inside-The-AI-Fueled-Romance-Scam-Explosion
Federal Bureau of Investigation, Internet Crime Complaint Center. “FBI IC3 2025 Internet Crime Report.” FBI, 2026. Summary via SOCRadar, https://socradar.io/blog/fbi-ic3-2025-internet-crime-report-10-takeaways/ and AARP, https://www.aarp.org/money/scams-fraud/fbi-ftc-report-2025-losses/
UK Finance. “Over £600 million stolen by fraudsters in first half of 2025.” UK Finance Press Release, 2025. https://www.ukfinance.org.uk/news-and-insight/press-release/over-ps600-million-stolen-fraudsters-in-first-half-2025
City of London Police. “A wrong turn on Love Lane: City of London Police take over city streets to warn of the dangers of romance fraud, with more than £106 million lost in the last year.” City of London Police, June 2025. https://www.cityoflondon.police.uk/news/city-of-london/news/2025/june/a-wrong-turn-on-love-lane-city-of-london-police-take-over-city-streets-to-warn-of-the-dangers-of-romance-fraud-with-more-than-106-million-lost-in-the-last-year/
TSB Bank / Fintech Global. “TSB warns romance fraud rising sharply across social media.” Fintech Global, 6 February 2026. https://fintech.global/2026/02/06/tsb-warns-romance-fraud-rising-sharply-across-social-media/
AARP. “Romance Scams Put Millions of Adults 50-plus at Risk: Here is What You Need to Know.” AARP Press Release, 3 February 2026. https://www.aarp.org/press/releases/2026-02-03-Romance-Scams-2026.html
Whitty, Monica T.; Buchanan, Tom. “The online dating romance scam: The psychological impact on victims, both financial and non-financial.” Criminology and Criminal Justice, 2016. https://journals.sagepub.com/doi/10.1177/1748895815603773
Operation Shamrock. “I Walked Into an Abandoned Scam Compound.” Operation Shamrock, 2026. https://www.operationshamrock.org/blog/erin-west-cambodia-inside-scam-compound
West, Erin. Written Testimony for the United States House Committee on Financial Services, REACT Task Force, September 2024. https://docs.house.gov/meetings/BA/BA10/20240918/117659/HHRG-118-BA10-Wstate-WestE-20240918.pdf
Springer Nature. “An Artificial Intelligence-Driven Temporal Network Analysis of Myanmar's Cyber Scam Ecosystem.” Asian Journal of Criminology, 2025. https://link.springer.com/article/10.1007/s11417-025-09474-0
American Bankers Association Banking Journal. “AI romance, 'machine-to-machine' scams among top 2026 fraud trends.” ABA Banking Journal, January 2026. https://bankingjournal.aba.com/2026/01/ai-romance-machine-to-machine-scams-among-top-2026-fraud-trends/
Norton. “What Norton's Insights Reveal About AI Dating Scams (2026).” Norton, 2026. https://us.norton.com/blog/online-scams/ai-online-dating-scams
Fox Business. “Tinder turns to eye scans to weed out bots amid rising concerns over romance scams.” Fox Business, 2026. https://www.foxbusiness.com/media/tinder-turns-eye-scans-weed-out-bots-amid-rising-concerns-over-romance-scams
UK Government. “Online Safety Act: Explainer.” GOV.UK, 2024. https://www.gov.uk/government/publications/online-safety-act-explainer/online-safety-act-explainer
Payment Systems Regulator. “APP scams reimbursement dashboard for Q3 2025.” PSR, 2025. https://www.psr.org.uk/information-for-consumers/app-scams-reimbursement-dashboard/
Electronic Frontier Foundation. “Ninth Circuit Correctly Rules That Dating App Isn't Liable for Matching Users.” EFF Deeplinks, February 2025. https://www.eff.org/deeplinks/2025/02/ninth-circuit-correctly-rules-dating-app-isnt-liable-matching-users
OpenAI. “Model Spec.” OpenAI, 18 December 2025. https://model-spec.openai.com/2025-12-18.html
OpenAI. “Findings from a pilot Anthropic-OpenAI alignment evaluation exercise: OpenAI Safety Tests.” OpenAI, 2025. https://openai.com/index/openai-anthropic-safety-evaluation/
United States Department of the Treasury. “U.S. and U.K. Take Largest Action Ever Targeting Cybercriminal Networks in Southeast Asia.” Treasury Press Release, October 2025. https://home.treasury.gov/news/press-releases/sb0278
Crypto.news. “Tether freezes $344M in USDT tied to sanctions evasion, pig-butchering scams.” Crypto.news, 2025. https://crypto.news/tether-freezes-344m-in-usdt-tied-to-sanctions-evasion-pig-butchering-scams/
NBC News. “How online romance scams are netting millions of dollars and pushing victims to self-harm.” NBC News. https://www.nbcnews.com/tech/tech-news/how-online-romance-scams-netting-millions-self-harm-rcna85252
ScienceDirect. “A qualitative investigation of the emotional, physiological, financial, and legal consequences of online romance scams in the United States.” Computers in Human Behavior Reports, 2024. https://www.sciencedirect.com/science/article/pii/S2949791424000605
Taylor and Francis Online. “Falling into a Black Hole: A Qualitative Exploration of the Lived Experiences of Cyberscam Victim-Survivors and Their Social Support Networks.” Victims and Offenders, 2025. https://www.tandfonline.com/doi/full/10.1080/15564886.2025.2481267
University of New South Wales. “Valentine's Day is coming: How vulnerable are you to AI-enabled romance scams?” UNSW Newsroom, February 2026. https://www.unsw.edu.au/newsroom/news/2026/02/valentine-s-day-is-coming--how-vulnerable-are-you-to-ai-scams
Coalition for Content Provenance and Authenticity. “C2PA Specification.” C2PA, 2024. https://c2pa.org/specifications/specifications/2.0/index.html

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast

Discuss...