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 Scale of the Problem: iGaming Fraud in 2025–2026
By the time 2026 arrived, online fraud in iGaming had outgrown every static defence built against it. TransUnion’s H1 2026 Fraud Trends Report found that 8.3% of attempted digital transactions globally in 2025 were suspected fraud – an 18% year-over-year increase that confirmed the arms race was still accelerating. Gaming platforms faced even steeper pressure: nearly one in ten US transactions was flagged as suspected digital fraud, and Gen Z users – the industry’s fastest-growing demographic – drove the sector’s highest fraud exposure at 11.7%.
Payment-method fraud reached 6.6% across all verticals, while multi-step coordinated fraud attacks – the kind that chain bonus abuse with synthetic identity creation and money laundering – rose by 180% year-on-year in 2025 according to the World Economic Forum’s December 2025 digital trust report. AI-assisted document forgery, which barely registered in 2024, accounted for 2% of all fake documents detected by Q4 2025 – a jump from near zero in twelve months.
The overall fraud rate did edge down from 2.6% to 2.2% across all sectors, proof that AI defences were making a measurable difference. But that headline figure masked a harder truth: fraud was getting more sophisticated, more automated, and more personalised at the same time that detection was improving. The operators who survived the squeeze were those who replaced rule engines with machine learning.
Market Size and Investment Trajectory
The AI-powered sports betting and iGaming fraud detection market was valued at USD 1,495.9 million in 2025, with projections pointing to USD 14,731.8 million by 2035 – a 25% CAGR that dwarfs almost every other enterprise software segment. Solutions (end-to-end platforms) accounted for 57.6% of that market, while services and point tools made up the rest. Cloud-based deployment led with an 84.8% share, reflecting the need for elastic, real-time fraud processing that can scale with live betting volumes during major sporting events.
Who Is Attacking and How
Fraud profiles shifted materially in 2025–2026. The World Economic Forum’s report documented the rise of generative AI synthetic identities – fabricated personas with plausible document images, realistic behavioural histories, and spoofed device fingerprints that could pass legacy KYC checks. Bonus abusers graduated from manual multi-accounting to bot farms running thousands of accounts in parallel, each with unique browser fingerprints generated by tools like GoLogin or Multilogin. Organised crime rings began using money mule networks coordinated through encrypted messaging channels, executing layered deposit-withdraw cycles that mimicked legitimate high-value players.
From Rules to Graphs: The Architectural Shift
The defining technical story of 2025–2026 was the transition from threshold-based rule engines to graph-based machine learning. Rule engines – “flag any withdrawal over $10,000 from an account less than 30 days old” – had served operators since online gambling began, but they were brittle. Fraud rings learned rule sets, built around them, and exploited the seams between systems.
Graph ML changed the calculus. Instead of evaluating each account in isolation, graph models map relationships across accounts, transactions, devices, and behavioural signals to expose hidden fraud rings. A device fingerprint shared between two ostensibly separate accounts is a weak signal in isolation; in a graph with 40 such connected devices, it becomes a high-confidence fraud cluster. This is the architecture that vendors like SEON, Sumsub, and Sardine moved toward in 2025, and it became industry baseline by mid-2026.
Behavioural Biometrics Enter the Stack
Alongside graph models, behavioural biometrics moved from experimental to production. By 2026, leading platforms were tracking cursor movement speed, typing cadence, swipe pressure, navigation flow, and session timing patterns to build per-user behavioural fingerprints. A returning player who suddenly types twice as fast and navigates directly to the withdrawal page without browsing games triggers anomaly detection – even if their credentials, device, and IP are identical to past sessions. Protegrity’s 2026 enterprise risk analysis cited behavioural biometrics as one of the three most impactful additions to the modern fraud stack alongside federated learning and graph ML.
Continuous Intelligence vs. Event-Based Checks
The other architectural shift was temporal. Legacy systems evaluated fraud at discrete events: account creation, deposit, bonus claim, withdrawal. By 2026, leading operators had moved to continuous behavioural intelligence – persistent monitoring that scores every session action in real time and updates a user’s risk profile dynamically. Protegrity’s 2026 analysis described this as the shift from “did this transaction cross a threshold?” to “does this user’s entire session trajectory match a known fraud pattern?” The difference in detection rates was substantial: false negatives that slipped through event-based checks were caught within minutes by continuous monitoring.
Key Vendors and Platform Milestones, 2025–2026
The vendor landscape consolidated rapidly. Below is a timeline of the most significant platform launches and upgrades during the period.
| Date | Vendor | Milestone | Key Feature |
|---|---|---|---|
| September 2025 | Mindway AI | GameScanner mobile launch | Mobile-first fraud scans – rapid betting detection, geo-location jump alerts |
| December 2025 | Genius Sports | ML fraud alerts integration | Historical betting pattern models for partner sportsbooks |
| January 2026 | Altenar | AI fraud suite upgrade | Live odds-shift checks, linked-account detection for sportsbooks |
| Q1 2026 | SEON | Graph ML v3 rollout | Multi-accounting ring detection across device + email + IP graph |
| Q1 2026 | Sumsub | AI document forgery module | GenAI-generated document detection; 98% accuracy on synthetic IDs |
| Q2 2026 | GeoComply | PinPoint v8 upgrade | Wi-Fi triangulation + VPN/proxy detection at <50ms latency |
“The 2024 fraud playbook was already obsolete by January 2025,” said Dr. Tamas Karalyos, Head of Risk Science at SEON, in a February 2026 interview. “Fraud rings were training their own ML models to probe our detection logic. Our response was to move the competition to the graph layer – where volume of connection signals matters more than any single feature.”
GeoComply, Sumsub, and the Compliance Convergence
One of 2026’s clearest trends was the convergence of fraud detection and regulatory compliance tooling into unified platforms. Historically, KYC (Know Your Customer), AML (Anti-Money Laundering), geolocation verification, and fraud detection were separate vendor contracts with separate dashboards. By mid-2026, the market was restructuring around integrated risk platforms that handled all four.
GeoComply’s PinPoint – used by over 700 licensed operators across 30 US states – added real-time fraud signal sharing in its v8 update, meaning a VPN spoofing attempt blocked in New Jersey automatically updated risk profiles across all integrated operator partners. For operators running cross-state operations, this network effect was significant: a fraud account blocked in one state was now pre-flagged before it could register in another. For a deeper look at GeoComply’s technical architecture, see our analysis of GeoComply vs Xpoint vs LocationSmart for US gambling geolocation.
Sumsub’s AI Document Forgery Module
Sumsub’s Q1 2026 release addressed the specific threat documented by the WEF: AI-generated fake identity documents. The module used a multi-layer CNN (convolutional neural network) trained on known GenAI document outputs – DALL-E, Midjourney, Stable Diffusion – to detect pixel-level inconsistencies in fonts, microprinting, and hologram patterns that human reviewers miss. Sumsub claimed 98% accuracy on synthetic ID detection in internal benchmarks, and a 0.3-second average check time that kept onboarding friction within acceptable UX bounds.
“The document forgery problem is fundamentally different from 2023,” said Vyacheslav Zholudev, CTO at Sumsub, in a March 2026 webinar. “In 2023, a fake ID was a physical scan of a modified real document. In 2026, it’s a fully synthetic image generated by a model that has learned what legitimate IDs look like. You cannot detect it with OCR or template matching – you need a model trained specifically on generative forgery artifacts.” For detailed KYC vendor benchmarks, the Sumsub vs Onfido vs Jumio vs Veriff iGaming KYC comparison covers pricing, speed, and accuracy across all four major platforms.
The AML Layer: Real-Time Transaction Monitoring
AML monitoring in iGaming shifted from batch-based (nightly review of flagged transactions) to real-time streaming analysis. Sumsub’s Travel Rule module, Sardine’s risk API, and ComplyAdvantage’s adverse-media feeds all moved to sub-second latency in 2025–2026. The practical implication: a deposit from a wallet linked to a newly sanctioned entity could be blocked before the transaction settled rather than caught in a next-day review queue.
Federated Learning and Privacy-Preserving AI
One of the less-publicised but technically significant shifts of 2025–2026 was the adoption of federated learning in multi-operator fraud intelligence networks. The challenge was structural: the most powerful fraud signals are cross-operator (the same fraud ring targeting multiple casinos simultaneously), but operators cannot share raw player data due to GDPR, CCPA, and gaming-specific data protection requirements.
Federated learning solved this by training shared fraud models on-device or on-operator-server, sharing only model gradient updates – not raw data – with a central aggregation server. The aggregated model improved on cross-operator signals without any individual operator exposing player PII. By Q2 2026, SEON and Sumsub had both launched federated learning consortiums for licensed EU operators, with early results showing a 23% improvement in fraud ring detection versus single-operator models.
Tokenisation and Synthetic Data Enrichment
Alongside federated learning, tokenisation and synthetic data enrichment became core privacy-preserving techniques. Operators generated synthetic player datasets – statistically indistinguishable from real data but containing no actual PII – for model training and third-party vendor testing. This allowed vendors to train and validate fraud models on realistic iGaming data volumes without regulatory exposure. Protegrity’s 2026 enterprise risk report identified synthetic data enrichment as a top-five investment priority for regulated gaming compliance teams. For broader market context, see our coverage of best sweepstakes casino sites.
Bonus Abuse Detection: From Heuristics to Prediction
Bonus abuse – creating multiple accounts to claim welcome offers, free spins, or reload bonuses – remained the highest-volume fraud category in iGaming through 2026. SEON’s 2025 iGaming fraud guide estimated that up to 30% of new account registrations at some operators were multi-account attempts, with bot farms capable of creating thousands of accounts per hour using residential proxies and browser fingerprint spoofing.
The AI response was predictive modelling at registration, not just post-claim detection. Modern platforms scored the probability of bonus abuse at account creation, using signals including: email domain age, social media footprint depth (or absence), device fingerprint rarity, IP velocity (how many other accounts used the same IP subnet in the past 30 days), and behavioural anomalies during the registration flow itself. Accounts scoring above a threshold were subject to enhanced KYC, reduced bonus eligibility, or a waiting period before activation.
Dynamic KYC for VIP Onboarding
The flip side of aggressive fraud detection is friction for legitimate high-value players. By 2026, the leading platforms had solved this with dynamic KYC: low-risk returning players completed standard verification in under 30 seconds, while new accounts with elevated risk scores were routed to enhanced due diligence. SEON’s data showed that operators using dynamic risk-tiered KYC reduced VIP player friction by 40% compared to uniform verification while simultaneously improving fraud catch rates.
What Competitors Don’t Cover: Beyond Standard Fraud Detection
Most iGaming fraud detection guides focus on the detection layer – flagging bad actors after they’re in your platform. What they consistently miss is the upstream fraud prevention ecosystem that leading operators deployed in 2025–2026:
- Pre-registration intelligence: Scoring fraud risk before account creation using email, phone, and IP signals – not just post-registration behavioural monitoring.
- Federated learning consortiums: Cross-operator model training that improves ring detection without sharing PII – rarely discussed in vendor marketing materials.
- GenAI document forgery detection: The specific CNN architectures trained to catch AI-generated fake IDs, which are fundamentally different from legacy document fraud.
- Crypto payment fraud: Lightning Network and on-chain transaction monitoring for crypto deposits, where traditional AML tooling doesn’t apply. For context on crypto payment providers in iGaming, our BitPay vs CoinsPaid vs CoinGate iGaming crypto payments comparison covers the monitoring capabilities of each.
- Regulatory divergence: How MGA (Malta), UKGC, and New Jersey DGE have different AI fraud detection requirements that affect vendor selection.
Regulatory Requirements: MGA, UKGC, and New Jersey DGE
AI fraud detection did not evolve in a regulatory vacuum. Three major licensing authorities issued updated guidance in 2025–2026 that directly shaped platform requirements.
The Malta Gaming Authority (MGA) published its 2025 Technical Standards Update requiring licensed operators to maintain real-time transaction monitoring with sub-5-minute suspicious transaction report (STR) generation. This effectively mandated streaming AML rather than batch processing for any MGA-licensed operator.
The UK Gambling Commission (UKGC) updated its AML and Fraud Prevention Guidance in November 2025 to require operators to document the specific AI models used in fraud detection, including training data sources and accuracy metrics, as part of their annual compliance submissions. This was the first regulatory requirement for AI model transparency in any major gambling jurisdiction.
The New Jersey Division of Gaming Enforcement (DGE) issued a March 2026 advisory requiring all licensed online operators to implement geolocation verification with VPN/proxy detection on every session start – not just at login – following a series of out-of-state betting incidents. GeoComply’s PinPoint became the de facto standard for DGE compliance.
| Regulator | Jurisdiction | Key AI Fraud Requirement (2025–2026) | Effective Date |
|---|---|---|---|
| Malta Gaming Authority (MGA) | EU / Malta | Real-time STR generation within 5 minutes | January 2026 |
| UK Gambling Commission (UKGC) | United Kingdom | AI model documentation in annual compliance filings | November 2025 |
| New Jersey DGE | New Jersey, USA | Per-session geolocation + VPN/proxy check | March 2026 |
| Ontario iGaming Ontario | Ontario, Canada | Real-time AML monitoring with 24hr STR filing | October 2025 |
| AGCO (Alcohol and Gaming Commission of Ontario) | Ontario, Canada | Mandatory KYC at C$500 lifetime deposit threshold | January 2026 |
The Compliance Cost Calculation
“Regulators aren’t just asking operators to detect fraud – they’re asking them to prove they can detect fraud, document how, and demonstrate that the model is current,” said Stacy Greenbaum, Gaming Compliance Partner at Greenberg Traurig, in a May 2026 industry panel. “That AI model transparency requirement from the UKGC is the most significant compliance shift I’ve seen in a decade. It means your fraud vendor is now part of your regulatory audit.” For broader market context, see our coverage of leading online casino sites.
The Economics of AI Fraud Detection
Pricing for AI fraud detection platforms varied significantly by deployment model and operator scale in 2025–2026.
SEON operated on a usage-based API pricing model starting at approximately $0.04–$0.10 per API call for standard device and email intelligence, with volume discounts for operators processing over 100,000 checks per month. Sumsub’s fraud module was bundled with KYC at $0.80–$3.80 per verified user depending on document type and jurisdiction, with AI fraud scoring included for enterprise contracts. GeoComply’s PinPoint was licensed on a per-check basis starting at $0.02–$0.05 per geolocation query for US operators.
The ROI case was compelling: a mid-sized online casino processing $50M monthly gross gaming revenue typically loses 1.5–2.5% to fraud without AI detection ($750K–$1.25M/month). A thorough AI fraud stack – SEON for device/email intelligence, Sumsub for KYC/document verification, GeoComply for geolocation – costs an estimated $15,000–$40,000/month at that scale, representing a 20–80x return on fraud losses prevented.
The False Positive Cost
“The metric that doesn’t get enough attention is false positive rate,” said Jamie Sherwood, Chief Risk Officer at a major European gaming group (speaking anonymously at ICE London 2026). “Every legitimate player you block costs you LTV, not just one transaction. At scale, reducing false positives by one percentage point is worth more than catching an extra one percent of fraud.”
By mid-2026, the leading platforms were achieving false positive rates below 0.5% on standard account checks, down from 2–3% in 2023. The improvement came from better feature engineering, larger training datasets, and the adoption of ensemble models that combined graph ML, behavioural biometrics, and device intelligence into a single risk score.
Crypto and Lightning Network Fraud Monitoring
As crypto payments grew in iGaming – driven by the Lightning Network’s near-instant settlements and sub-cent fees – fraud monitoring had to expand beyond card-and-bank-transfer frameworks. Lightning Network transactions don’t have the same on-chain trail as standard Bitcoin transactions, creating a monitoring gap that traditional AML tools couldn’t address.
CoinsPaid’s Crystal Intelligence integration and Chainalysis’s Reactor tool emerged as the leading solutions for on-chain transaction monitoring in 2025–2026. Crystal Intelligence provided real-time risk scoring for wallet addresses, flagging addresses associated with known mixing services, darknet markets, or sanctioned entities before deposits settled. Chainalysis Reactor traced the full transaction graph across chains, with specific Lightning Network path analysis added in its Q3 2025 update.
Expert Predictions: What AI Fraud Detection Looks Like by 2028
With the 2025–2026 evolution as a baseline, the industry’s leading voices converged on several forward projections for the remainder of the decade.
- Fully autonomous fraud response: By 2027, the leading platforms will close the loop between detection and response – automatically adjusting deposit limits, triggering enhanced KYC, and routing high-risk sessions to human review without operator intervention. Dr. Karalyos of SEON predicted this “autonomous risk management” would be standard in licensed markets by H1 2027.
- Cross-jurisdictional fraud intelligence networks: Federated learning consortiums will expand from single-jurisdiction pilots to multi-market networks. SEON and Sumsub both signalled plans for transatlantic consortiums by 2027, sharing model updates across EU and US licensed operators.
- Regulatory AI audits: The UKGC’s model documentation requirement will be adopted by the MGA and New Jersey DGE by 2027, creating a global standard for AI fraud model disclosure that vendors will need to support natively.
- GenAI red-teaming: Operators will begin using adversarial AI – generative models trained to simulate fraud attempts – to continuously probe and improve their detection systems. Early pilots were running in Q2 2026 at several major European operators.
- Biometric liveness as standard: Active liveness checks (video selfie, challenge-response) will become table stakes for all new account creation in regulated markets by 2028, eliminating static photo-based KYC entirely.
“The fraud detection category in iGaming will look nothing like a rule engine by 2028,” said Vyacheslav Zholudev of Sumsub. “It will be an autonomous, self-updating risk intelligence system that learns from every fraud attempt across every operator in real time. The human element will shift entirely to oversight and exception management.”
Frequently Asked Questions
What is AI fraud detection in iGaming?
AI fraud detection in iGaming uses machine learning models – including graph ML, behavioural biometrics, and neural networks – to identify fraudulent account activity, bonus abuse, money laundering, and payment fraud in real time, replacing or augmenting static rule engines.
Which vendors lead the iGaming AI fraud detection market in 2026?
SEON, Sumsub, GeoComply, Mindway AI, Genius Sports, and Altenar are among the leading vendors as of mid-2026, each specialising in different parts of the fraud stack: device intelligence, KYC, geolocation, responsible gambling signals, and sportsbook integrity respectively.
How much does AI fraud detection cost for an iGaming operator?
Pricing varies by vendor and volume. SEON charges approximately $0.04–$0.10 per API call, Sumsub bundles fraud scoring with KYC at $0.80–$3.80 per verified user, and GeoComply’s PinPoint starts at $0.02–$0.05 per geolocation check. A thorough stack for a mid-sized operator typically costs $15,000–$40,000/month.
What is federated learning in iGaming fraud detection?
Federated learning allows multiple operators to collaboratively train shared fraud detection models without sharing raw player data. Each operator trains the model locally and shares only gradient updates with a central server, preserving GDPR/CCPA compliance while benefiting from cross-operator fraud signals.
What did the UKGC require for AI fraud detection in 2025?
The UK Gambling Commission updated its AML and Fraud Prevention Guidance in November 2025 to require licensed operators to document the specific AI models used in fraud detection – including training data sources and accuracy metrics – as part of annual compliance submissions.
How does AI detect bonus abuse in iGaming?
AI models score bonus abuse risk at account registration using signals including email domain age, device fingerprint rarity, IP velocity, social media footprint depth, and behavioural anomalies during the sign-up flow. Accounts above a risk threshold are subject to enhanced KYC or reduced bonus eligibility.
What is the false positive rate for leading AI fraud detection platforms in 2026?
Leading platforms achieved false positive rates below 0.5% on standard account checks by mid-2026, down from 2–3% in 2023. The improvement came from ensemble models combining graph ML, behavioural biometrics, and device intelligence into a unified risk score.
How does AI fraud detection handle crypto payments in iGaming?
Crypto AML monitoring in iGaming uses tools like Crystal Intelligence (CoinsPaid integration) and Chainalysis Reactor to score wallet addresses in real time and trace transaction graphs across chains. Chainalysis added Lightning Network path analysis in its Q3 2025 update to address the monitoring gap created by Lightning’s off-chain settlement model.
