AI fraud detection in iGaming did not arrive quietly in 2026 – it arrived as an arms race. Through the first half of the year, online gambling operators found themselves defending against coordinated fleets of autonomous fraud agents while regulators handed out eight-figure fines and a market that barely existed five years ago crossed the billion-dollar line. Global identity fraud rates actually fell from 2.6% in 2024 to 2.2% in 2025, yet iGaming operators reported more fraud than ever. That paradox – fewer attacks, far more damage – is the story of how fraud defense evolved through 2026, and it reshapes everything from KYC onboarding to how much operators budget for their risk teams.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
The 2026 Inflection Point: When iGaming Fraud Became an AI-vs-AI War
For most of the last decade, iGaming fraud was a volume game. Bots hammered signup forms, bonus hunters spun up throwaway accounts, and rule-based systems swatted them down with velocity checks and IP blocklists. That era ended in 2025. According to Sumsub’s Identity Fraud Report 2025-2026, the share of multi-step attacks – chains that link synthetic-identity creation, deepfake verification, bonus abuse and money laundering into a single automated flow – jumped from 10% of all identity fraud in 2024 to 28% in 2025. That is a 180% year-on-year increase in “sophisticated” fraud, a figure independently echoed by the World Economic Forum’s December 2025 analysis.
From 2.6% to 2.2% – the headline that hides the real story
The global identity fraud rate dropping to 2.2% looks like a win, and in narrow terms it is: stronger verification made cheap, high-volume attacks unprofitable. But the decline masks a redistribution of risk. Simple attempts collapsed while sophisticated, AI-driven operations exploded. Fraudsters stopped throwing thousands of dumb bots at the wall and started deploying a handful of very smart ones. The result is fewer incidents that each cost dramatically more to detect, investigate and remediate.
Why “sophistication” replaced “volume”
Sumsub co-founder and CEO Andrew Sever framed the shift bluntly: “The Sophistication Shift marks a turning point, as businesses now face challenges tied to their velocity: the speed at which they can detect threats and adapt.” In practice, velocity is now the entire game. A fraud ring that can generate a fake ID, pass a deepfake liveness check and mimic human betting behaviour in seconds forces operators to make risk decisions in milliseconds – a demand no rule table can satisfy.
A $1.5 Billion Market Racing Toward $14.7 Billion
The economics of defense have followed the threat. The market for AI-powered fraud detection across sports betting and iGaming was valued at US$1,495.9 million in 2025, according to Future Market Insights, and is forecast to reach US$14,731.8 million by 2035 – a compound annual growth rate close to 25%. On that trajectory, the segment is on track to clear roughly US$1.9 billion in 2026 alone.
Where the money goes: end-to-end platforms and cloud
Two structural facts define how operators spend. End-to-end platforms – suites that bundle onboarding KYC, transaction monitoring, device intelligence and case management – account for 57.6% of spend, as operators consolidate point solutions into single vendors. And cloud deployment dominates at 84.8%, reflecting an industry that needs to push model updates continuously rather than ship quarterly software releases. On the ground, Tech-Insider’s June 2026 vendor-pricing survey found mid-sized operators typically spending US$15,000–US$40,000 a month on their combined AI fraud stack, with unit costs that vary sharply by vendor and layer: SEON charges roughly US$0.04–US$0.10 per API call for a fraud check, while Sumsub prices KYC plus AI fraud scoring at US$0.80–US$3.80 per verified user – a gap that reflects the difference between a lightweight signal check and full identity verification.
The 2026 spend trajectory
That consolidation is not academic. When an operator can lose a £10 million fine over AML failures, the calculus shifts from “cheapest tool” to “most defensible stack.” Buyers increasingly want vendors who can demonstrate model governance, auditability and explainability to a regulator – capabilities that favour larger, better-capitalised platforms over scrappy startups.
| Metric | Figure | Period | Source |
|---|---|---|---|
| iGaming & sports-betting AI fraud-detection market | US$1,495.9M | 2025 | Future Market Insights |
| Same market (projected) | ~US$1.9B | 2026 | Future Market Insights (25% CAGR) |
| Same market (forecast) | US$14,731.8M | 2035 | Future Market Insights |
| Global identity fraud rate | 2.2% (from 2.6%) | 2025 | Sumsub / WEF |
| iGaming-specific fraud rate | 1.53% (+18% YoY) | Q1 2026 | Sumsub |
| Multi-step attack share of ID fraud | 28% (from 10%) | 2025 | Sumsub |
| Deepfake fraud growth | +700% | Q1 2024 → Q1 2025 | Sumsub |
| Suspicious-transaction volume | 4.5× | Q1 2025 → Q1 2026 | Sumsub |
| Operators reporting more fraud | 82.9% | 2025 | Sumsub |
The Numbers Behind the 2025-2026 Fraud Surge
Zoom into iGaming specifically and the “declining fraud” narrative collapses entirely. While the cross-industry rate fell, the iGaming-specific fraud rate rose to 1.53% in Q1 2026 – an 18% year-on-year increase and a 40% jump from 2024. That verification-level figure likely understates the operational burden: a separate Tech-Insider analysis published in June 2026 put the broader flagged-transaction rate – covering suspicious wagers, deposits and withdrawals across online casino and sportsbook traffic, not just failed identity checks – at 8.3%. Gambling is uniquely exposed because it combines instant money movement, generous promotions and a global, often pseudonymous user base – exactly the surface autonomous fraud agents are built to exploit.
iGaming’s 1.53% rate and the 4.5× transaction spike
Sumsub recorded a 4.5× increase in suspicious transaction volumes for iGaming between Q1 2025 and Q1 2026 – a figure PlayUSA’s June 2026 reporting on Sumsub’s data corroborated, alongside the same outlet’s finding that fraudulent verification attempts hit 1.53% in Q1 2026, up 18% year-on-year – while 82.9% of operators surveyed said fraud had increased over the prior year. Payment-method fraud became the dominant threat vector in 2025, surpassing document fraud for the first time as attackers pivoted from faking identities to compromising the money rails themselves.
Deepfakes at 700% and 11% of first-party fraud
The most alarming curve is synthetic media. Deepfake fraud rose 700% globally between Q1 2024 and Q1 2025 and now accounts for roughly 11% of first-party fraud. In Europe, deepfakes make up 41% of all fraud attempts – the highest share of any region. Synthetic identity document fraud, meanwhile, climbed 195% globally (and 300% in the United States), as generative tools made a convincing fake passport a commodity rather than a craft.
The Technology Evolution: Rules → ML → Graph ML → Agentic
To understand how defense caught up, you have to trace four distinct eras of detection technology – a progression that competitor explainers rarely lay out chronologically.
Static rules to supervised ML (2023-2024)
The starting point was hand-written rules: block this IP range, flag deposits above a threshold, reject accounts sharing a device. Cheap and explainable, but brittle – fraudsters simply probed the boundaries until they found the gaps. Supervised machine learning risk scoring, which dominated 2023-2024, was a leap forward: models learned fraud patterns from labelled historical data. But supervised models share a fatal flaw in this domain – they need examples of fraud they have already seen, leaving them blind to novel attack chains.
Graph ML becomes the 2026 baseline
By mid-2026, graph machine learning had become the industry baseline. Instead of scoring accounts in isolation, graph ML maps relationships across accounts, transactions, devices and behavioural signals to expose hidden fraud rings – the connective tissue that individual risk scores miss. Vendors including SEON, Sumsub and Sardine rebuilt their architectures around graph analytics precisely because multi-accounting and “gnoming” (bonus abuse via networks of synthetic identities) are relationship problems, not single-account problems. Sportsbook-specific integrity tools followed the same trajectory: Genius Sports rolled out machine-learning fraud-alert models for sportsbook operators in December 2025, scoring wagers against historical betting patterns to flag anomalous activity that a single-account risk score would never catch.
Continuous behavioral intelligence and behavioral biometrics
The newest layer is continuous behavioural intelligence: persistent monitoring that scores every session action in real time and updates a player’s risk profile dynamically rather than at fixed checkpoints. NuxGame reported in July 2026 that operators running these always-on AI fraud stacks were holding event-detection latency under five seconds even at wallet-and-rewards scale, a threshold that keeps scoring fast enough to intervene mid-session rather than after the fact. Mindway AI pushed this layer onto mobile with the September 2025 launch of its GameScanner module, built specifically for rapid, on-device betting-fraud detection. Behavioural biometrics – how a user types, swipes, holds a device and navigates a betting slip – now supplements document checks, because behaviour is far harder for a generative model to fake convincingly than a face.
| Era | Dominant approach | Strength | Blind spot |
|---|---|---|---|
| Pre-2023 | Static rules / thresholds | Cheap, explainable | Rigid; fraudsters test around limits |
| 2023-2024 | Supervised ML risk scoring | Learns known patterns | Needs labelled data; misses novel rings |
| 2025 | Graph ML + device intelligence | Maps hidden relationships and rings | Compute-heavy; data silos |
| 2026 | Agentic AI + continuous behavioural intelligence | Real-time, adaptive, session-level | Governance, explainability, cost |
Agentic AI: The Threat That Defines 2026
If graph ML was the defining defensive shift, agentic AI is the defining offensive one. Sumsub’s November 2025 annual report was titled, tellingly, “Fraud Shifts to Complex Multi-Step Schemes in 2025, Agentic AI Scams Poised to Surge in 2026.” The prediction is already playing out.
Fleets of autonomous fraud agents
An AI fraud agent combines generative models with reinforcement learning to create synthetic identities and interact with verification systems in real time. In some documented cases, a single agent can orchestrate an entire attack chain – generating a fake ID document, submitting a deepfake video for liveness, and conducting human-like interaction at machine speed. Analysts now warn of coordinated fleets of these agents conducting high-speed, multi-step attacks at a scale capable of overwhelming traditional anti-fraud systems.
When defenders and attackers both go agentic
The response has been symmetrical. Sumsub launched AI Agent verification – a system to verify AI agents that increasingly transact on behalf of real users – while Sardine invested its 2025 Series C directly into agentic defense. This is the genuine 2026 inflection: the front line is no longer human analysts versus scripts, but defensive AI agents versus offensive AI agents, with humans supervising the exchange. For operators building compliance stacks, robust iGaming KYC verification platforms are the first and most important line where this agentic clash actually happens.
The SEON Paradox: AI Everywhere, Yet Fraud Teams Keep Growing
Here is the finding that upends the conventional wisdom that AI would shrink compliance headcount. SEON’s “AI Reality Check: 2026 Fraud & AML Leaders Report,” published on 24 February 2026 and based on a survey of 1,010 fraud, risk and compliance leaders, found that 98% of organisations now use AI in fraud and AML workflows and 95% are confident it works – yet headcount plans rose from 88% to 94% year-on-year, and 83% expect budgets to increase in 2026.
98% adoption, 94% still hiring
In betting and gaming specifically, 57% of operators report fraud outpacing revenue growth. AI did not replace analysts; it changed what they do. The bottleneck moved from detection to orchestration – connecting fragmented data, tuning models, investigating the harder cases that automation escalates, and defending decisions to regulators.
Tamas Kadar: “integration as strategy, not plumbing”
SEON co-founder and CEO Tamas Kadar captured the paradox directly: “Fraud and financial crime were supposed to become more manageable as AI matured. Instead, 2026 is the year leaders are confronting a more complicated reality. AI adoption is real, confidence is high, but the scale and pace of fraud – compounded by fragmented systems – continue to drive increased investment rather than reduced overhead.” His prescription: “The bottleneck is no longer whether AI works. It’s everything around it: disconnected data, siloed teams, slow implementations. The organizations that pull ahead will be the ones that unify fraud and AML intelligence, shorten the distance between threats and controls, and treat integration as strategy, not plumbing.”
Deepfake KYC and the Death of “Blink and Turn”
Nowhere has the arms race been more visible than at the onboarding gate. Active liveness checks – the familiar “blink, turn your head, smile” prompts – were designed to defeat static photo spoofs. Against real-time deepfake injection, they became a liability, because a generative model can render a blink or a head-turn on demand. For broader market context, see our coverage of best sweepstakes casino sites.
Passive liveness and behavioral biometrics
Operators responded in 2025-2026 by replacing active challenges with passive liveness detection, which analyses micro-textures, lighting physics and camera artefacts invisible to the user, paired with behavioural biometrics captured during the session. The shift is subtle for legitimate players – verification for low-risk returning users now completes in under 30 seconds – but brutal for injected synthetic media.
Dynamic KYC and delayed friction
The organising principle is “dynamic KYC” or delayed customer friction: route low-risk returning players through frictionless verification, and escalate high-risk new accounts to enhanced due diligence only when signals warrant it. This is where onboarding meets money movement, and why operators increasingly pair identity checks with real-time AML transaction monitoring so that a clean signup can still be caught if its later transactions light up. A simplified real-time risk decision now looks less like a rulebook and more like a scored, multi-signal routing gate:
# Illustrative real-time risk decision (2026-style routing)
signals = {
"device_linked_accounts": graph.count_linked_accounts(device_id),
"session_velocity": behavior.actions_per_minute(session_id),
"passive_liveness_score": biometrics.passive_liveness(selfie),
"deposit_withdrawal_gap_s": ledger.gap_seconds(user_id),
}
risk = model.score(signals) # gradient-boosted + graph embeddings
if risk > 0.85 or signals["device_linked_accounts"] > 5:
route = "enhanced_due_diligence" # step-up KYC
elif risk < 0.20:
route = "frictionless" # verify in < 30s
else:
route = "manual_review" # human analystHow the Losses Break Down: ATO, Bonus Abuse, Multi-Accounting
Fraud in iGaming is not monolithic, and 2026 data finally quantified where the damage concentrates. SEON’s 2026 report found that three categories account for 68% of all betting-and-gaming losses.
The 27/23/18 split
Account takeover (ATO) leads at 27% of losses, followed by promotion and bonus abuse at 23% and loyalty-program abuse at 18%. The pattern is instructive: two of the top three are not “break-in” fraud at all but abuse of the operator’s own incentive machinery – exactly the surface synthetic identities are built to farm.
Gnoming and synthetic-identity bonus farms
Group-IB, whose Head of Gaming Sarah Psaila has spent more than two decades on the practitioner side of iGaming fraud, has documented the rise of “synthetic players”: AI-generated personas that mimic legitimate behaviour to harvest sign-up bonuses at scale, laundering the proceeds across networks of linked accounts. This is why graph ML matters so much – a lone synthetic account looks clean, but the ring it belongs to does not. Operators running provably fair mechanics still need these controls; even a mathematically certified RNG and audited game integrity does nothing to stop a coordinated bonus-farming ring at the identity layer.
| Fraud type (betting & gaming) | Share of losses |
|---|---|
| Account takeover (ATO) | 27% |
| Promotion / bonus abuse | 23% |
| Loyalty-program abuse | 18% |
| Combined top three | 68% |
| Other (chargebacks, collusion, laundering) | 32% |
Vendor Consolidation: Visa, Feedzai, Sardine and the Platform War
The defensive market did not just grow in 2025-2026 – it consolidated, as payment networks and fraud specialists raced to own the full stack.
Visa’s $1.1B Featurespace bet
Visa completed its acquisition of Featurespace in December 2024 for a reported US$1.1 billion, folding the adaptive-behavioural-analytics firm into its Risk and Identity Solutions unit. Featurespace’s technology protects around 500 million consumers and processes over 100 billion payment events per year – reach that now sits directly inside the card rails iGaming operators depend on.
Sardine’s agentic pivot and Soups Ranjan
Sardine raised a US$70 million Series C in February 2025 – a raise Film Daily’s July 2026 coverage credited with fueling the expansion of its agentic fraud platform into casinos – and poured it into agentic AI. CEO and co-founder Soups Ranjan – previously Head of Financial Crime at Revolut and Head of Risk at Coinbase – described the pressure that is driving automation: “Risk teams are stretched to their limits, and the workload keeps growing. Alert volumes have surged 800%, compliance hiring can’t keep up, and analysts are stuck in an endless cycle of manual reviews. Sardine’s AI agents automate the most time-consuming parts of risk operations, helping teams move faster, reduce false positives, and focus on stopping real threats.”
Feedzai and the digital euro
Feedzai, which raised US$75 million at a US$2 billion valuation, was selected by the European Central Bank as a first-ranked provider for fraud and risk management for the digital euro, slated for launch by 2029. That endorsement matters for iGaming because the same real-time scoring engines vetted for central-bank digital currency are the ones operators license for gambling payments.
The Regulators Bite: £10m Fines and the A$5,000 Threshold
Technology evolved because the cost of failure did. Between May and December 2025, the UK Gambling Commission took regulatory action against 13 operators for AML and social-responsibility failings – and the details read like a warning label for lazy automation. For broader market context, see our coverage of leading online casino sites.
The UKGC’s 13-operator crackdown
The headline penalty was £10 million against Platinum Gaming in October 2025. But the most instructive case was Videoslots’ £650,000 fine, where the Commission explicitly criticised an “over-reliance on an algorithm to monitor risk” that let a customer move the proceeds of gambling across multiple accounts without adequate source-of-funds checks. The message: an AI model is not a compliance defence if you cannot show a human was watching it. Operators paying processors to move money still carry the AML liability, which is why the choice of licensed gambling payment processors is now as much a fraud-governance decision as a cost one.
AUSTRAC, Star Entertainment, and the 31 March 2026 reform
Australia’s AUSTRAC pressed even harder. Star Entertainment Group faces a potential AUD 400 million (about US$260 million) penalty after a June 2025 hearing over AML/CTF non-compliance – including allegedly extending a AUD 266.7 million cheque-cashing facility to a customer with known organised-crime links. From 31 March 2026, Australia’s reformed AML/CTF regime lowers the gambling customer-due-diligence exemption threshold from A$10,000 to A$5,000 and strengthens ongoing monitoring, dragging far more routine play into scope.
| Regulator | Operator | Penalty | Date | Basis |
|---|---|---|---|---|
| UKGC | Platinum Gaming | £10,000,000 | Oct 2025 | AML / SR failings |
| UKGC | Spreadex | £2,022,000 | 7 May 2025 | AML / SR |
| UKGC | ProgressPlay | £1,000,000 | 21 Aug 2025 | AML breaches |
| UKGC | Done Brothers (Betfred) | £825,000 | 3 Dec 2025 | AML / SR |
| UKGC | Corbett Bookmakers | £686,070 | 2025 | AML / SR |
| UKGC | Videoslots | £650,000 | 2025 | Over-reliance on monitoring algorithm |
| UKGC | Maple (Lottomart) | £360,000 | 17 Sep 2025 | AML settlement |
| AUSTRAC | Star Entertainment | up to A$400M (proposed) | Hearing 4 Jun 2025 | AML / CTF |
| AUSTRAC | Entain (Ladbrokes/Neds) | Civil proceedings | 16 Dec 2024 | AML / CTF |
Federated Learning and Consortium Intelligence
One structural answer to sophisticated, cross-operator fraud rings emerged in 2026: sharing intelligence without sharing data. Because a fraud ring rarely attacks a single operator, defenders that see only their own traffic are always at a disadvantage.
Sharing signals without sharing data
Federated learning lets multiple operators train a shared fraud model on their collective patterns while each keeps its raw customer data local – the model travels, the data does not. For licensed platforms bound by GDPR and gambling-specific data rules, this is the difference between collaborative defence and a privacy breach. Early consortium approaches focus on identifying device and identity fingerprints that recur across operators, exposing rings that any single platform would miss.
The privacy and regulatory tightrope
The catch is governance. Sharing even derived signals across competing operators raises competition-law and data-protection questions that legal teams are still working through in 2026. Expect consortium intelligence to grow fastest in tightly regulated markets where operators already share a compliance framework, and slowest where cross-border data transfer rules remain unsettled. Geolocation integrity adds another shared layer, since US gambling geolocation compliance signals frequently expose the same spoofing infrastructure fraud rings reuse across state lines.
Beyond the Standard Coverage: What SEON, Group-IB and CrossClassify Don’t Tell You
The most-read explainers on iGaming fraud – from SEON’s definitional guide to Group-IB’s generative-AI blog and CrossClassify’s product FAQ – are useful, but they share three blind spots this analysis deliberately fills.
No timeline, no market data, no regulatory teeth
First, vendor explainers describe fraud types (multi-accounting, gnoming, bonus abuse) without a chronological account of how defense evolved from static rules to agentic AI – the four-era progression that actually explains 2026. Second, they omit hard market data: the US$1,495.9 million 2025 market size, the 57.6% platform-spend share, the 84.8% cloud-deployment figure. Third, and most importantly, they rarely connect the technology to the regulatory consequences – the £10 million Platinum Gaming fine or the A$5,000 AUSTRAC threshold that make this a boardroom issue, not just a risk-team one.
The vendor-consolidation story they skip
Product pages naturally avoid discussing their competitors’ M&A. But Visa’s US$1.1 billion Featurespace acquisition, Sardine’s US$70 million agentic pivot and Feedzai’s digital-euro selection are the real market signals telling operators where defensive capability is concentrating. Understanding who owns the stack you are buying into is as important as the feature checklist.
Five Predictions for AI Fraud Detection in iGaming Through 2027
Extrapolating from the verified 2025-2026 trajectory, five developments look most likely.
- Agentic-vs-agentic becomes standard. By 2027, defensive AI agents that autonomously investigate and action alerts will be table stakes, not differentiation – the human role shifts fully to supervision and exception handling.
- The market clears US$2.5 billion. At the observed ~25% CAGR, iGaming fraud-detection spend passes US$1.9B in 2026 and approaches US$2.5B in 2027, with consolidation continuing as payment networks acquire specialists.
- Passive liveness fully replaces active checks. As real-time deepfake injection matures, “blink and turn” prompts disappear from serious onboarding flows entirely in favour of passive detection and behavioural biometrics.
- Federated consortia go mainstream in regulated markets. Shared-signal networks become a competitive necessity in the UK, EU and Australia, even as competition and privacy regulators scrutinise them.
- Headcount keeps rising, not falling. SEON’s paradox persists: AI raises analyst productivity but the pace of sophisticated fraud keeps net compliance headcount growing through 2027.
What This Means for Operators and Players
For operators, the strategic takeaway is that fraud defense is now a governance discipline as much as a technical one. Buying the best model is necessary but not sufficient; regulators fine you for how you supervise it, integrate it and document it. The winners in 2026 are the operators who unified fragmented fraud and AML data, shortened the distance between a detected threat and an enforced control, and can prove a human owns every automated decision.
For players, the trade-offs are real but improving. Legitimate users increasingly pass through frictionless, sub-30-second verification, while the friction concentrates on genuinely high-risk signals. The uncomfortable truth beneath the reassuring 2.2% headline is that the fraud that remains is smarter, faster and more expensive to stop than anything the industry faced before – and through 2026, defenders are, for the first time, fighting it with the same weapons.
Frequently Asked Questions
What is AI fraud detection in iGaming?
It is the use of machine learning – increasingly graph ML and agentic AI – to identify and stop fraud on online gambling platforms in real time. It spans KYC/onboarding checks, transaction monitoring, device intelligence, behavioural biometrics and case management, replacing the static rule-based systems that dominated the industry before 2023.
How big is the iGaming AI fraud detection market in 2026?
The market for AI-powered fraud detection across iGaming and sports betting was valued at US$1,495.9 million in 2025 and is on track to clear roughly US$1.9 billion in 2026 at a ~25% CAGR, according to Future Market Insights, reaching a forecast US$14,731.8 million by 2035.
What is the biggest AI fraud threat in iGaming in 2026?
Agentic AI. Autonomous fraud agents that combine generative models with reinforcement learning can generate a fake ID, pass a deepfake liveness check and mimic human betting behaviour in a single automated chain. Multi-step attacks rose from 10% of identity fraud in 2024 to 28% in 2025 – a 180% year-on-year increase.
How much did deepfake fraud rise?
Deepfake fraud increased 700% globally between Q1 2024 and Q1 2025 and now accounts for roughly 11% of first-party fraud. In Europe, deepfakes make up 41% of all fraud attempts – the highest share of any region – which is why operators replaced active “blink and turn” liveness checks with passive detection.
Why are iGaming fraud teams still growing if AI works?
SEON’s 2026 Fraud & AML Leaders Report found 98% of organisations use AI and 95% are confident it works, yet headcount plans rose from 88% to 94% year-on-year. AI raised analyst productivity but did not keep pace with the scale and sophistication of fraud, so operators are investing in both technology and people.
What are the main types of iGaming fraud losses?
According to SEON’s 2026 data, account takeover accounts for 27% of betting-and-gaming losses, promotion and bonus abuse for 23%, and loyalty-program abuse for 18% – together 68% of all losses. Much of it is abuse of operators’ own incentive systems rather than external break-ins.
What regulatory actions hit iGaming operators in 2025-2026?
The UK Gambling Commission sanctioned 13 operators between May and December 2025, including a £10 million penalty for Platinum Gaming and a £650,000 fine for Videoslots over “over-reliance on an algorithm.” In Australia, Star Entertainment faces a potential A$400 million AUSTRAC penalty, and a reformed AML/CTF regime took effect on 31 March 2026.
What is graph ML and why does it matter for fraud detection?
Graph machine learning models relationships between accounts, devices, transactions and behaviours rather than scoring each account in isolation. It matters because modern iGaming fraud is a network problem – a single synthetic account looks legitimate, but the ring of linked accounts it belongs to reveals the fraud. Graph ML became the industry baseline by mid-2026.
