Writing by Amir Shachar.https://amirshachar.com/enhttps://amirshachar.com/blog/alert-budgets-not-auc/https://amirshachar.com/blog/alert-budgets-not-auc/Fraud is rare and review capacity is fixed, so judge a fraud model by the frauds in the alerts a team can open and by the money each decision moves.Thu, 12 Nov 2026 12:00:00 GMTfraudevaluationhttps://amirshachar.com/blog/fraud-rings-and-mule-networks/https://amirshachar.com/blog/fraud-rings-and-mule-networks/Ring payments look normal one at a time, so the signal is in shared devices, beneficiaries and timing, and a graph score is a lead to confirm.Thu, 12 Nov 2026 12:00:00 GMTfraudgraphsmuleshttps://amirshachar.com/blog/propose-then-verify/https://amirshachar.com/blog/propose-then-verify/A language model can propose fraud rules and explanations, but a deterministic check, separate from the model, should decide what runs in a payment flow.Thu, 12 Nov 2026 12:00:00 GMTfraudgenerative-aimodel-riskhttps://amirshachar.com/blog/the-label-you-get/https://amirshachar.com/blog/the-label-you-get/Fraud labels arrive late and only for transactions someone approved, reviewed or disputed, so training splits and evaluation have to account for that.Thu, 12 Nov 2026 12:00:00 GMTfraudlabelsevaluation