Financial Services
Banking, lending and financial crime: foundation models trained on bank data, same-day credit decisions, and fraud defences from dark-web intelligence to formally verified rules.
Fraud & financial crime
7 articlesFour ways to stop fraud earlier: upstream dark-web intelligence, real-time graph detection, provably correct rules and synthetic fraud data.
Fraud & financial crime
Across 7 articles: find the threat before it reaches the bank, detect it in under 100 ms, enforce rules that are proven correct, and train on data you would otherwise not have.
- Dark Web Fraud Signals3-part series · Turning dark-web chatter into early-warning signals for bank fraud models.
- Real-time Fraud with TGNs2-part series · Real-time graph fraud detection with sub-100 ms inference that generalises across payment domains.
- How we built a formally verified fraud rules engine on AWSBuilder Center · A fraud rule that is wrong, or can be bypassed, costs money and creates regulatory exposure.
- Augment fraud transactions using synthetic data in Amazon SageMakerAWS Blog · Fraud models lack enough high-quality fraud examples to train on.
Dark web fraud signals for banking anti-fraud models (Part 1): why upstream intelligence changes the game
The problem: Fraud techniques show up on dark web forums weeks before they reach bank transaction monitoring.
Dark web fraud signals (Part 2): building the intelligence pipeline
The problem: How do you collect and classify dark web content safely and at scale?
Dark web fraud signals (Part 3): detection rules, composite alerts, and deployment
The problem: How do you turn raw intelligence into alerts a fraud team can act on?
Detecting fraud in real time with Temporal Graph Networks on AWS (Part 1)
The problem: Fraud rings move faster than batch models can detect them.
Detecting fraud in real time with Temporal Graph Networks on AWS (Part 2)
The problem: A fraud model trained on one payment domain does not generalise to the next.
How we built a formally verified fraud rules engine on AWS
The problem: A fraud rule that is wrong, or can be bypassed, costs money and creates regulatory exposure.
Augment fraud transactions using synthetic data in Amazon SageMaker
The problem: Fraud models lack enough high-quality fraud examples to train on.
Lending & decisioning
2 articlesCutting loan decisions from days to minutes, with audit built in.
The Fast Path: how any bank can cut loan decisions from days to minutes
The problem: Bank loan decisions take about 10 days, and 95% of that time is spent chasing data rather than modelling.
The Fast Path, Built: a reference architecture for minutes-not-days bank decisions
The problem: What does the architecture for same-day lending actually look like?
Bank foundation models
4 articlesWhy and how a bank trains its own foundation model, from privacy-first data to Nova Forge.
Why banks should train their own foundation models (Part 1 of 3)
The problem: General-purpose models do not understand a bank's own event data.
Building the data foundations for a bank foundation model (Part 2 of 3)
The problem: How do you build training data from customer events without breaching privacy?
Training and deploying a bank foundation model on AWS (Part 3 of 3)
The problem: How do you take a bank foundation model from training to production?
Extending your bank foundation model with Amazon Nova Forge (Part 4)
The problem: How do you add language understanding to a structured banking model?
Governance, risk & sovereignty
2 articlesMaking AI decisions auditable and in-region for regulators.
Sovereignty is not one thing: why your regulated NOC needs more than a region
The problem: "Just keep it in-region" treats sovereignty as a checkbox.
Mechanical governance for LLM decisions
The problem: Regulated banks need AI decisions they can audit, and LLMs are probabilistic.
More in Financial Services
1 articleSeries in this industry
The Autonomous NOC
A self-improving AI reasoning layer that turns an alarm storm into one root cause in under 90 seconds, safely enough to act on a live network.
The Fast Path
Cutting bank loan decisions from days to minutes in six weeks.
Bank Foundation Models
Why and how a bank should train a foundation model on its own event data, from privacy-first data to Nova Forge.
Dark Web Fraud Signals
Turning dark-web chatter into early-warning signals for bank fraud models.
Real-time Fraud with TGNs
Real-time graph fraud detection with sub-100 ms inference that generalises across payment domains.
Synthetic Data
Measuring and using synthetic data where real data is restricted.