Tag
The author built an MCP tool that lets AI agents retrieve a Google Maps reviewer's full history and metadata for vetting reviews and detecting fraudulent patterns.
SeqLLM is a framework that injects behavioral-sequence modeling into pretrained LLMs while preserving language ability, enabling joint analysis of text and behavior for high-stakes decisions. Deployed at WeChat Pay, it improves merchant screening precision from 92.0% to 97.5% and achieves state-of-the-art results on open recommendation benchmarks.
This paper introduces a hierarchical human-AI triage model for POS fraud detection in Nigerian FinTech, designed to neutralize structural bias that discriminates against rural agents due to infrastructure-related noise. The system uses a calibrated ensemble model, specialist analysts, and senior supervisors to achieve substantive equality of opportunity, reducing the regional performance gap from 19.43 to 2.88 percentage points.
RUBRIC is a generator-agnostic filtering framework for imbalanced classification that selects synthetic samples by balancing realism (via a discriminator) and utility (margin-based scoring), improving F1-macro and recall on benchmarks like credit-card fraud detection.
This paper presents an exploratory benchmark for detecting sparse-ring fraud in dynamic transaction graphs using quantum-inspired Contextual Machine Learning (CML) compared to a GRU baseline, finding that hybrid graph features combining identity-preserving and topological summaries yield the best results.
A comment on how the decision not to forward 'sort by amount received' to fraud departments is always a policy choice, highlighting organizational priorities.
An interactive exploration of Benford's Law across real datasets, explaining the mathematical phenomenon where the digit 1 appears as the first digit about 30% of the time, and its applications in fraud detection.
This paper presents a controlled comparison of seven categorical encoding methods on the IEEE-CIS fraud dataset, evaluating accuracy, interpretability, and computational trade-offs with a fixed LightGBM learner, finding that entity embeddings achieve the highest AUC-ROC but no encoder dominates both metrics.
NVIDIA's blog post describes an end-to-end workflow using PyTorch-native NeMo AutoModel for pretraining a transaction foundation model. The workflow uses GPU-accelerated data processing and tokenization, decoder-only model pretraining, and embedding extraction to improve fraud classification performance by over 40% on the IBM TabFormer dataset.
This paper presents a synthetic multimodal framework for insurance fraud detection at the first notice of loss (FNOL). It generates dialogue transcripts and two-speaker audio, combining ASR, NER, LLM-RAG, and speaker embeddings into a rule-based risk scoring system.
Proposes the Multi-Stream Fraud Transformer (MSFT) for financial fraud detection, which independently encodes transaction, login, and risk event streams using Transformers and fuses them with time-aware positional encoding and gated fusion, achieving 0.9961 AUROC on a large dataset.
Proposes TMR-GGNN, a time-aware multi-relational graph neural network for credit card fraud detection that handles imbalanced data and evolving fraud patterns via contrastive learning and focal loss.
This article from Sumsub examines the rising threat of AI-powered deepfake fraud targeting the creator economy, and offers a guide on detection and protection methods to combat impersonation and voice cloning scams.
A practitioner recounts how deploying AI agents in production required 80% engineering effort on workflow, ownership, and approval processes rather than the model itself, highlighting that the 'boring layer' of shared context and routing is critical for real-world impact.
Introduces SAGE, the first end-to-end LLM-driven multi-agent framework for fraud detection, using a Data Diagnostic Tree and Markov decision process with natural-language gradients to optimize models under class imbalance. Experiments show significant F1 improvements over baselines across five datasets.
Silicon Valley titan Peter Thiel shares PayPal's early experience of being on the verge of bankruptcy due to fraud, and how it got out of trouble through human-machine collaboration (computer screening + human qualitative investigation), pointing out that this collaborative paradigm is underestimated by the AI research community.
Financial institutions are shifting from siloed AI models to unified transaction foundation models built on transformer architectures, as demonstrated by NVIDIA's report and Revolut's PRAGMA model, which improves fraud detection, credit scoring, and recommendations while reducing feature engineering effort.
SilIF augments Isolation Forest with a silhouette-based scoring layer on per-tree path length fingerprints, improving unsupervised transaction fraud detection on the IEEE-CIS benchmark by +0.0080 AUC-PR on average.
A merchant details a case of friendly fraud where Stripe refused to use clear evidence of abuse to improve its cross-merchant fraud detection, despite marketing its Radar product on network-wide signals.
This survey examines computational nondeterminism in financial AI systems, covering tabular models, graph networks, and LLM-based workflows, and proposes a layered evaluation framework for auditability.