Comprehensive, accurate, and ethical AI with global reach.
Stay ahead of evolving fraud with advanced AI technology
The document fraud landscape never stands still. The number of global identity documents is always growing, and so is the attack surface available to fraud. Document verification solutions have to scale to support thousands of global document types, each with unique shapes, sizes, and security features. The solution must also be able to detect a range of fraudulent attack methods.
Meeting these needs requires deep up-to-date expertise and fast adaptation to new attacks. Static approaches can restrict your growth and leave you vulnerable to emerging threats. The most dangerous fraud out there is the fraud you haven’t yet seen.
10,000+ micro-models trained to detect specific fraud attack vectors with precision.
How does Atlas AI keep you two steps ahead of fraud?
Built from the ground up over the last 10 years by a team of 200+ engineers and researchers, Atlas provides a scalable, accurate, and consistent layer of defense from known fraud attack vectors, and reacts fast to protect your business against never-before-seen fraud.
Micro-model architecture to keep your business safe
Built in-house, Atlas is based on our learnings from processing tens of millions of documents globally. It’s made of thousands of micro-models that are trained to look for even the slightest shifts in color, texture, shape, and specific fraud signals on individual documents. This means an ID is analyzed by machine-learning models that are experts on that document. And if fraud does get through, we retrain specific micro models with just a single sample. That’s how we provide unmatched accuracy and speed at scale to protect your business, no matter where in the world your customers are.
Agile reactions to stop tomorrow’s attack
Atlas is expertly trained to recognise what fraudulent and genuine documents look like. This means it’s able to spot known fraud attacks with precision, reducing missed fraud cases by 54% compared with machine learning approaches trained only to spot fraud they’ve seen before. Training Atlas on genuine documents as well as fraudulent documents means that when new attack vectors arise, we accurately detect anomalies and retrain the models in days, not months, empowering you to react quickly.
54% reduction in missed fraud, compared with machine learning approaches using general models trained only to spot fraud they’ve seen before.
Always testing, always teaching
We’ve created our own Fraud Lab to ensure that we’re improving our product against the latest fraud trends and techniques. Using our rich datasets and proprietary technology, we train our models using the best insights available, to protect your business today and in the future.
How it works
1. Atlas classifies the document against a database of over 2,500 supported document templates, each expertly categorised and annotated with up-to-date fraud insights by our in-house Document Specialist teams.
2. It then deploys micro-models tuned specifically to the different components of that document, detecting anomalies in font, texture, colour, and security features, to assess for fraud. Each model is trained on insights gathered from our expansive proprietary datasets of genuine and fraudulent documents.
3. We then supply both high-level and granular results through API and the Onfido Dashboard. By detailing up to 38 breakdowns on why a Document passed or failed review you can automate decisioning to optimize your onboarding processes.
Learn more about how we drive value in our customer case studies.