Privately, ACCS, Idiap and AVPA join forces to protect age estimation from generative AI
Generative AI services are emerging that can potentially be used to thwart online facial age estimation systems, just as those systems are mandated in a growing number of jurisdictions.
A collaboration between the Age Check Certification Scheme (ACCS), Idiap Research Institute, the Age Verification Providers Association (AVPA) and Privately SA has been announced to develop tools to defend against spoof attacks with generative AI.
The project, referred to by Privately in a LinkedIn post as “Safeguarding Age Estimation and Digital IDs from Emerging Threats Posed by Generative AI,” is funded by a grant from Innovate UK and Innosuisse.
“I think we can all agree, Age Assurance systems are going to be pretty useless if a 12 year old can simply spoof them with an easily available App,” explains ACCS Executive Director Tony Allen in a separate LinkedIn post. “In order to maintain confidence in the efficacy of the systems, we need a suite of test assets (sometimes called a test harness) and agreed protocols and standards for applying those tests.”
Privately brings its experience with biometric age and ID verification systems, while Idiap will contribute its highly-regarded biometrics research expertise. The ACCS is the steward of the UK’s age estimation standards, and the AVPA is the industry’s global advocacy and support organization.
The roadmap for the project begins with comprehensive research, and will proceed to “deliver tangible and immediate benefits” for the age estimation sector.
ID R&D CSO and Co-founder Konstantin Simonchik notes the threat injection attacks pose to biometric security and the lack of established standards, asking “how can we ascertain that the injection tests conducted are thorough and comprehensive?”
The collaborative research project intends to provide an answer to this pressing question, Allen says.
Article Topics
Age Check Certification Scheme (ACCS) | age estimation | AVPA | biometrics | face biometrics | generative AI | Idiap | Privately | spoof detection
Comments