Ant International has taken first place in the NeurIPS Competition of Fairness in AI Face Detection, a contest focused on building facial recognition models that work accurately across gender, age, and skin tone. The win comes as the company pushes to strengthen security and fairness across its global payments network, especially as deepfake technology becomes more common.
Facial recognition now plays a growing role in everything from customer onboarding to payment verification. But research from the U.S. National Institute of Standards and Technology has shown that many widely used systems still struggle to identify women and people of color, often due to narrow training datasets. Those gaps can lead to security weaknesses and, in some cases, the denial of essential financial services.
This year’s NeurIPS competition drew more than 2,100 submissions from 162 teams worldwide. Participants were tasked with detecting 1.2 million AI-generated face images representing a wide range of demographic groups. Ant International’s entry stood out for its approach, which blends a Mixture of Experts architecture with a built-in bias-detection mechanism.
The team trained two neural networks at the same time: one that focuses on spotting deepfakes and a second that challenges the first by pushing it to ignore demographic cues. This setup forces the system to focus on authentic signs of manipulation rather than traits linked to age, gender, or skin tone. The model also drew on a globally representative dataset and real payment fraud cases to help ensure it could scale.
“A biased AI system is inherently an insecure one,” said Dr. Tianyi Zhang, general manager of risk management and cybersecurity at Ant International.
He noted that building fairness into the model is essential for spotting deepfakes and maintaining reliable identity checks for users everywhere.
Ant International is now integrating the winning technology into its financial services and says it can detect deepfakes with more than 99.8% accuracy across all demographics in the 200 markets where it operates. The company serves more than 150 million merchants and 1.8 billion user accounts through services such as Alipay+, Antom, Bettr, and WorldFirst.
The model also supports global eKYC standards used during customer onboarding, which is especially important in regions where limited access to inclusive financial tools remains a challenge. Ant’s broader risk-management framework, AI SHIELD, underpins these efforts. Built on an AI Security Docker system, it is designed to reduce vulnerabilities such as unauthorized access and data leaks. The company says its Alipay+ EasySafePay 360 tools have already lowered account takeover incidents in digital wallet payments by 90%.
