Tag
An arXiv paper formulates payroll fraud detection as inductive link prediction on synthetic HR/ERP graphs, comparing analyst rules, path features, and GNNs. A gradient-boosted model on chained-join features outperformed a GraIL-style GNN on link AP, while the relational entity GNN led under failed identifier joins, showing multi-hop gains are not GNN-specific.
This paper introduces SITO, a synthetic independent benchmark of 2,332 contrastive cases in English, Ukrainian, and Surzhyk, used to audit the Tuteliq Detect commercial text-safety API for missed harm and false positives across six harm categories, finding notably weaker recall for third-party bystander reports of ongoing harm.
该arXiv论文提出面向代理式商务（Agentic Commerce）的行为感知欺诈检测与网络安全框架，用于实现可信的AI代理支付授权，并评估了其攻击检测率与分类性能指标。
PayPal AI introduces TxnQA, a benchmark for analytical QA over merchant transaction event sequences, covering 27 task families and a fraud-risk reasoning suite. Evaluation of proprietary and LoRA-adapted open-weight models reveals schema compliance, arithmetic calculation, and evidence attribution as three distinct bottlenecks.
RGH-GNN 提出一种风险门控的异构图神经网络，将 LightGBM 生成的交易级风险分数作为边权重引导消息传递，以提升金融欺诈检测性能，在 IEEE-CIS 数据集上取得 0.914 的 AUC-ROC 并超越所有基线。
Zoom and major financial firms are adopting World's 'Proof of Human' badges, powered by iris-scan verification, to combat deepfakes and AI-driven fraud in sensitive conversations.
澳大利亚初创公司 Apate.ai 部署数千个 AI 代理伪装成受害者与骗子互动，已浪费骗子 17 年时间并收集欺诈情报，供银行、电信和政府机构反欺诈使用。该公司是本届 TechCrunch Disrupt Startup Battlefield 200 的决赛入围者。
Google DeepMind与新加坡GovTech合作，基于开源模型Gemma微调构建设备端诈骗检测系统，可在保护个人隐私的同时实时识别本地社交工程诈骗威胁。
本文提出 TRACE 治理框架，用于量化与追踪生产 AI 系统中的「可解释性债务」，并通过为期十二个月的欺诈检测案例研究展示其预测审计风险的能力，为欧盟 AI 法案第 13 条合规提供首个可量化运营架构。
提出了一个混合量子-经典框架（QML-CRL），通过参数化量子电路编码认知偏差，并结合强化学习以最优停止问题建模，用于检测亲密关系及恋爱投资欺诈。
This arXiv paper introduces a role-aware Decider-Supervisor framework combining centralized ML, a FedAvg-trained federated meta-model, and Base/QLoRA LLM variants with blockchain auditability, finding that AI supervision only reduces fraud-decision errors under specific roles, calibration, and traffic conditions rather than by default.
SR-Fraud is an outcome-supervised reflective LLM agent framework for non-stationary payment fraud detection, improving detection metrics over traditional methods on a production benchmark.
CRISP is a scalable coreset method for imbalanced tabular learning that efficiently reduces dataset size while maintaining high accuracy in tasks like fraud detection.
This paper proposes Centroid-Guided Contrastive Loss (CGCL) for structured fraudulent job posting detection, unifying classification and clustering in latent space to achieve state-of-the-art performance.
This paper introduces StreamFraudNet, a weakly supervised incremental model for detecting phone scams from raw speech, achieving a ROC-AUC of 0.9953 and operating in real-time to provide early warnings during calls.
The article reports on AI-powered dating app scams uncovered by Anthropic, where autonomous AI personas impersonate real people to defraud users.
The article argues that many enterprise AI pilots fail because they use generative models for tasks better suited to discriminative models, highlighting differences in mathematical objectives and update mechanisms.
FraudBench is an executable benchmark for stress-testing policy-grounded banking agents against adaptive fraud attacks, simulating scenarios with tools, mutable state, and internal policy documents.
Stripe's acquisition of OpenRouter is focused on ecosystem-wide AI security and alignment, using OpenRouter's AI traffic data to improve fraud detection and risk management through continuous context feedback loops.
This paper studies the application of AI and machine learning algorithms for recognizing fraudulent banking transactions, proposing preprocessing techniques and comparing models. An artificial neural network and stacked generalization achieve improved AUC scores, with the best result around 0.954.