* Uses various sophisticated heuristics to detect real vs fake IDs.
* Matches the ID to the human face.
* Detects whether the human face is live or not.
* Dynamically requests more or less information depending on the confidence level.
It also gets better over time based on the attacks and fraud attempts that Stripe itself sees.
Those are 4 great bullets btw. They helped me understand the service a lot more than the landing page for identity. Might want to consider a view like this.
We’re exploring different options for scanning IDs like Anyline and BlinkID right now, but this looks incredibly well suited for what we’re building and would save us a tremendous amount of time if it works.
• Document detection
• Blur and glare detection
• Tool-tips during the user flow
This means it will fail for a few individuals and you will be stuck trying to reach support who are going to be pointless and useless.
Mostly this will be an issue with people of non-white ethnic origins and people with older laptops/phones with poorer cameras.
This is obviously a useful product, but it is one the world would be much better of if it didn't exist at all.
> Document checks verify the authenticity of government-issued identity documents. Stripe uses a combination of machine learning models, automated heuristic analysis and manual reviewers to verify the authenticity of hundreds of different document types.
> Selfie checks look for distinguishing biological traits, such as face geometry, from a photo ID and a picture of your user’s face. Stripe then uses advanced machine learning algorithms to ensure the face pictures belong to the same person.
> ID Number checks provide a way to verify a user’s name, date of birth, and national ID number. Stripe uses a combination of third-party data sources such as credit agencies or bureaus, utility or government-issued databases and others to verify the provided ID number.
> Match the ID photo with selfies of the document holder
> Validate SSN and addresses against global databases
Seems fairly clear.