PYMNTS报告：诈骗占欺诈金额22.4%，银行调整防御策略
PYMNTS Intelligence基于2026年7月对90家美国银行、信用合作社和FinTech高管的调查发现，诈骗占欺诈金额的22.4%，高于2025年的12%；诈骗与账户接管合计占50.6%，两年前为27.6%。
#AI in Payments
Today 0 itemsPYMNTSSource summaryAI score5353 PYMNTSSource summaryAI score5757 FIS与Anthropic推出面向美国银行的反洗钱调查智能体
FIS与Anthropic开发的FIS AML Investigations Agent现已面向美国银行商业化提供，用于反洗钱调查。该智能体汇集核心系统、支付和数据中的证据，结合客户类型、历史活动与预期行为分析警报，并解释判断依据。Amalgamated Bank已部署该智能体，嵌入式控制让分析师保留每项决定权。
PYMNTSSource summaryAI score6161 英格兰银行警告银行警惕 AI 网络安全风险
英格兰银行副行长 Sarah Breeden 表示，AI 的进展已使其对全球金融系统构成系统性风险，银行不能假设网络攻击者无法突破防线。英格兰银行正在调查银行应对量子计算加密风险的准备情况，并计划据此制定覆盖金融行业的系统性应对方案。Breeden 还强调，银行需要为核心服务准备独立于主系统的 bare metal 恢复方案，并尽早迁移至抗量子加密方法。
PYMNTSSource summaryAI score4242 AI 成为英国金融监管体系的一部分：FCA部署工具审查文件并识别金融犯罪威胁
英国金融行为监管局（FCA）正将 AI 纳入金融监管流程，用于审查机构文件、加快常规案件处理并从分散数据中识别金融犯罪威胁。FCA称，AI自动化已将简单案件处理时间从最多四小时降至平均约六分钟，但监管判断和决策仍由人工负责。FCA还完成了 Palantir 为期12周的试点，用于整合多类数据，结果及后续合作仍待评估。
PYMNTSSource summaryAI score5353 PYMNTS Intelligence：AI欺诈防御使用者报告更高的付款前拦截率
PYMNTS Intelligence与Bottomline合作的报告显示，使用AI欺诈防御的企业中，77%称其效果优于人工方法，59%表示可在资金离开企业前阻止至少90%的欺诈尝试；未使用AI工具的企业对应比例为32%。AI采用者中83%使用实时预测风险评分，82%使用自动文件验证，78%扫描收到的邮件和短信以识别AI生成迹象。尚未使用AI欺诈检测的企业中，58%计划在未来12个月内实施。
CryptopolitanSource summarySelectedAI score7777 Japan’s FSA asks banks and crypto asset exchanges to stop using identity document photo verification early
Japan’s Financial Services Agency (FSA) has notified banks and crypto asset exchanges not to wait until April 1, 2027, but to stop requiring customers to upload photos of identity documents early and switch to verification by reading data from identity card IC chips. Under the new process, smartphones read the customer’s name, address, date of birth and facial photo, and match them against a live image of the applicant; the facial image capture step remains.
推荐理由：The material explains Japan’s shift from identity document photo uploads to IC chip-based verification and notes that banks and crypto asset exchanges need to adjust their remote account opening processes in tandem.
PYMNTSSource summaryAI score5151 PYMNTS报告：59%采用3种以上AI工具的企业可在损失前拦截至少90%的支付欺诈攻击
PYMNTS Intelligence报告显示，在采用3种以上AI工具防范支付欺诈的企业中，59%能在损失发生前拦截至少90%的攻击；没有AI防御的企业中这一比例为32%。报告基于对美国150名年收入至少为$100 million的企业财资和财务高管的7月调查，83%的AI使用者采用real-time risk scoring，82%使用自动化文件验证。
Payments DiveSource summaryAI score5252 Certos调查显示企业金融负责人正应对AI欺诈与深度伪造
Certos对230名高级金融负责人开展的调查显示，81%的受访企业过去一年遭遇过涉及或疑似涉及AI生成、增强内容或技术的欺诈尝试。84%的受访者认为AI欺诈比传统欺诈更难识别，只有约一半有信心在资金离开组织前发现相关欺诈或骗局；43%最担心付款转移，93%表示AI欺诈风险正在改变其对金融机构合作伙伴的看法。
CryptopolitanSource summarySelectedAI score8181 South Korea’s Financial Services Commission suspends plan to relax bank security rules
South Korea’s Financial Services Commission has suspended its plan to relax bank security rules after AI-assisted attacks affected 7 financial institutions and exposed tens of thousands of customer records. The next round of participating banks was originally scheduled to be selected on October 7. The plan concerns South Korea’s network isolation rules. In the first round, 49 institutions with at least 10 trillion won in assets and at least 1,000 employees received one-year exemptions, while the second phase was originally intended to expand the program to 75 companies.
推荐理由：The material connects regulatory relaxation, AI penetration-testing tools and data leaks at multiple institutions, helping readers understand the security and testing trade-offs facing the network isolation policy.
Plaid BlogSource summaryAI score6060 Plaid推出 Fraud Foundation Model，升级 Plaid Protect 欺诈防范
Plaid推出 Fraud Foundation Model，将行为与交易序列数据预训练后，再根据客户特定的欺诈标签进行微调，并接入 Trust Index 现有评分框架。该模型使用数亿条专有数据点训练，内部测试显示其欺诈检测表现较此前 Trust Index 模型提升40%，目标是更早识别欺诈并减少人工审核和正常用户摩擦。
PYMNTSSource summaryAI score4040 银行重审身份核验：AI 伪造声音与面孔
PYMNTS Intelligence 与 Paymentology 联合报告显示，2025 年未经授权方欺诈占欺诈事件和金额损失的 71%，高于上年的 48%，欺诈正从孤立交易转向基于身份的持续性攻击。AI 正削弱语音、视觉和行为生物特征核验的可靠性，银行需转向持续、基于风险的验证，并以带消费限额、商户限制的令牌化替代暴露支付凭证，以应对 AI 智能体代客下单带来的风险。
FCA (UK)Source summarySelectedAI score6262 FCA outlines regulatory direction for next-generation market infrastructure: from experimentation to scaled adoption
FCA Chief Executive Nikhil Rathi said at a TheCityUK dinner that the UK must move from experimentation to adoption and create the conditions for emerging technologies such as tokenization to scale safely in wholesale markets. FCA has authorized HSBC Orion to provide real-time digital securities custody services, plans to finalize the stablecoin regime before summer, and will jointly publish a tokenization roadmap with the Bank of England while consulting on protection rules for related tokenized investment assets.
推荐理由：FCA has clearly stated its intention to move from experimentation to scaled adoption and provided timelines for tokenization, stablecoins, and digital securities custody, allowing readers to assess the regulatory direction of the UK’s wholesale market infrastructure.
Fintech NewsSource summaryAI score3333 Ripjar 在 ULTRA 中升级 AI 客户筛查能力，打击全球犯罪经济
Ripjar 为其 ULTRA 引擎引入新 AI 功能，该引擎服务于 25% 的全球系统重要性银行及 35 家以上全球财富 500 强企业。ULTRA 每日解析超 100 亿条历史文章和 500 万条新文章，覆盖 150 多种语言，风险归因准确率达 98.7%，将到达分析师的警报量减少 99%。
FINTRAC (Canada)Source summaryAI score5454 加拿大金融业FIFAI II报告提出AGILE框架应对AI风险与机遇
FIFAI II最终报告梳理AI对加拿大金融业带来的风险与机遇，并提出AGILE框架，即Awareness、Guardrails、Innovation、Learning和Ecosystem Resiliency。