Shahid HanifShufti’s CEO, and co-founder, has extensive technology experience. She is an entrepreneur who has built identity verification platforms, fintechs, blockchains, and software decentralized platforms. He spent seven years in the role of Chief Technology officer, where he led in-house research and development for Shufti’s artificial intelligence-driven document and biometric verification technology. Hanif founded Developers Studio in 2017, a company that develops blockchain technology with over 100 experts. He was also CTO at Quickbit where he developed the cryptocurrency payment system ahead of its initial public offering. He co-founded Programmers Force, and expanded the company’s software and data sciences to over 500 employees in ten offices across three continents.
Shufti It is an identity verification platform powered by artificial intelligence that can help organizations build trust, reduce fraud and comply with Know Your Business and Know Your Customers requirements. The platform combines document and biometric validation, electronic identification verification, NFC checks, business and age verification, fraud detection and ongoing monitoring through one unified infrastructure. It supports more than 150 different languages and thousands of documents in over 240 nations and territories. This allows businesses to confirm customers and organisations through a global integration. This technology uses a combination of biometric validity, device intelligence (device intelligence), and document authenticity to determine if forged documents are being used, or if account manipulation is taking place.
If you are a co-founder Shufti You were Chief Technology officer in 2017 before you became CEO in the year 2024. You built the platform to address the shortcomings of digital identification verification. How has this changed as generative AI becomes more prevalent?
We co-founded Shufti with a friend in 2017. The biggest challenge was slow verification. There were too many manual tasks and the systems didn’t function well between countries. Most identity verification systems were inconsistant, particularly for industries with high risk. Also, they struggled with documents written in non-Latin language and were unable to reliably confirm identities across the globe.
Today’s generative AI is changing the game. The problem is no longer limited to reading ID documents. The key is to verify that the ID document, and the individual presenting it, are genuine. AI made identity fraud easier, faster and cheaper. We have learned that a simple verification during the onboarding process is not enough. AI is needed to detect the latest fake identities and documents.
Belgian authorities recently warned In the last 12 months, more than 10 000 people have been victims of AI enabled identity fraud. This case reveals how rapidly identity fraud is evolving. Why are AI-generated versions of legal documents so difficult to detect and what does it reveal about the speed at which this crime has evolved?
Belgian authorities’ warning about 10,000 victims was just the beginning. The report reveals criminals are moving from simple document editing to full scale identity synthesis.
AI-generated copys are hard to detect as they can imitate government templates at pixel level precision. OCR (Optical character recognition) focuses on extracting the text but ignores the integrity of the images. AI is now able to replicate the security features previously requiring physical presence for verification, making flat images of documents a liability instead of proof of identity.
What is the difference between an AI-generated version of a stolen document and a forgery conventionally done, a document that has been manipulated, or a synthetic identity created from scratch?
These methods are important to differentiate.
Conventional Forgery: A physical counterfeit document.
Manipulated ID: This is a legitimate document where certain fields have been changed.
Synthetic identity: A “Frankenstein” Persona created by mixing real information (such as an SSN number) and fabricated data.
AI-Generated copy: Deepfake documents created using a template stolen or from scratch, and generated by Generative Adversarial Networks. These documents often do not have a digital history or contain artifacts, such as inconsistencies with sensor noise, that the human eyes cannot detect.
Shufti predicts that deepfakes of documents could rise by 3,900% in this year. What is the activity that drives this projection? And what assumptions and limitations should be understood by organizations when they interpret it.
AI makes fraud easier to produce and scale. We anticipate a 3900% rise in deepfakes. The criminals no longer swap faces on photos. Criminals can produce fake documents containing all the information needed to prove their identity.
This projection is important because it shows how fast AI fraud has grown, and not how many fakes documents there are. It’s important to understand that this projection reflects how quickly AI fraud is growing, not just the number of fake documents. Fake identities are able to pass through the checks, and stay hidden within company databases.
A photograph or scan is still considered sufficient as proof of identification by many businesses. What additional signals would a modern system of verification examine beyond visible data on the document itself?
Businesses can no longer rely solely on document verification. What we call an “Digital Eye” approach. In order to determine whether an ID is authentic, it must be examined by the system.
It means looking out for small signs, that are difficult to spot by humans, such as if the image has been captured directly from a computer screen or not, unusual pixel patterns and lighting variations, or any traces of AI-generated images. In addition, we look for evidence that the document has been digitally edited, copied or moved. We also examine metadata and image quality. By combining all of these indicators, it’s much easier to spot sophisticated AI generated documents than older verification methods.
Your argument is that the assurance of identity should not be limited to onboarding customers, but rather continuous. How would the continuous verification process look in reality? And how could companies use it to avoid excessive surveillance or privacy concerns, as well as customer friction and frustration?
This is the age of the “one-time check” The end is near. The Continuous Identity Assurance system refreshes user risk every 15 seconds against the 1,700+ watchlists to help prevent. “retroactive non-compliance.”
FastID is a biometrically bound reusable identity that allows us to implement the system without friction and privacy concerns. After a user has been verified, he or she can verify high-risk activities (like large withdrawals), in less than two seconds with only a face scan. The need for document uploads is eliminated while still maintaining the highest level of security.
Now, fraudsters are able to combine fake documents, face swaps and deepfake videos with injection attacks. They can also use stolen information. What should the identity platform do to connect biometrics, document integrity, device intelligence and behaviour analysis in order to detect these coordinated attacks.
Fraudsters have become more and more sophisticated. To bypass identity verification, fraudsters now use a combination of AI-generated documentation, face swaps, injection attacks, and AI generated documents. It’s important that businesses look at all aspects of the situation rather than just relying solely on one step.
Shufti does this by using context-aware scorers. We use iBeta Level 2 passive liveness certification to verify the physical presence of the user. Then we analyze identity data and device fingerprints and examine user behavior. These signals can be combined to help identify frauds before they cause damage.
The generational models will improve, but fraudsters may compress, rescan or modify synthetic media in order to mask manipulation artifacts. How can verification providers determine if their systems are still effective in detecting previously unknown attack methods?
AI-generated fraud moves much faster than security testing cycles. The verification providers must constantly evaluate their systems in relation to new attacks, rather than just relying upon historical data. This means that you should test with synthetic documents and manipulated media to conceal obvious artifacts. As generative AI advances, it is becoming more difficult for hackers to duplicate certain types of deepfakes.
If identity verification systems incorrectly refuse legitimate users, this can lead to serious consequences. What are the best ways to measure fraud detection accuracy, false positives and demographic performance?
Accuracy isn’t about just catching the target. “bad guys”It is about creating a frictionless journey for everyone. “good guys.” In our industry, we focus on the Failure to Extract Rate (FTXR)—which measures how often a system simply fails to “read” a face or document—and the False Non-Match Rate (FNMR), where genuine users are incorrectly rejected. Shufti performed at the DHS RIVR benchmark 2025 with 0% of extractions failing across multiple devices, and a worst case FNMR under 0.68%.
The developer must go beyond “lab averages” Measure performance by “worst-case” Results by demographics This includes testing for uniformity in skin tones and facial structures as well as cultural clothing. This is achieved by educating our AI with global datasets that contain millions of frames. It’s not complete unless you can prove that the system works as well in rural areas with low lighting conditions as in an office setting. A Unified Identity Layer should be fair, inclusive and easily accessible for all.
In the near future, will technology such as digital identity wallets supported by governments, credentials that can be verified cryptographically, and identities bound to biometrics render uploaded documents obsolete? Or will these technologies simply provide a whole new surface for attack?
The future is one where we won’t have to take photos of our identity documents so often. The government-backed digital wallets for identity and digital IDs are more efficient and secure than images of documents because the information is verified directly. The time taken to perform the verification is also reduced, which makes for a better user experience.
Criminals are always looking for new ways to exploit new technologies. Attackers may instead try to steal digital credentials or take control of accounts. Digital credentials are not enough. The person presenting the ID must be verified as the owner. This can be done through biometrics and by checking the liveness of the individual. In the future, identity verification will likely combine digital credentials and biometrics to create multiple layers of security instead of just relying solely on one method.
Please visit our website for more information about this great interview. Shufti.

