Workshop showcases EU progress on remote identity verification, but fragmentation persists
The legal, technical and user requirements for remote digital identity verification are sometimes conflicting, but need to be managed by service providers and other stakeholders, and to this end ENISA and ETSI have organized a joint workshop on remote identity verification and the ecosystem that the eIDAS regulation addresses.
Prof. Dr. Rainer Herpers from the Institute for Visual Computing at the University of Bonn-Rhein-Sieg opened the event with a presentation on deep attacks against identity verification systems.
The following presentation, by Battista Biggio, PhD., of the Patterns Recognition & Applications Lab at the University of Cagliari and co-founder of Pluribus One, explored the use of adversarial attacks on machine learning systems through disturbances at the pixel level or when labeling training data.
The combination of deepfakes and adversarial attacks could pose a particular threat to remote digital identity verification systems, according to the panelists.
ENISA’s report on attacks and countermeasures was also briefly presented.
Juliette Delanoe, co-founder and CMO of Ubble.ai, said the company’s research shows that an average of 5-6% of digital identity verification attempts it can pass judgment on are fraudulent. . She also provided a breakdown of the frequency of types of fraud.
That number was slightly different for the other panelists representing the biometrics industry in the first session, IDnow founder and chief technology officer Armin Bauer and Veriff co-founder and CPO Janer Gorohhov.
Gorohhov said Veriff has found a fraud rate of six to eight percent, depending on the industry served, and up to 10 percent in cryptocurrency. Social engineering is the most common attack vector observed by IDnow, reports Bauer.
All agreed that deepfakes are an imminent attack vector, but not uncommon today.
Asked about the impact of NFC and electronic IDs on document fraud, Delanoe argued that the tools that are effective today can be complemented by NFC, but will not be replaced by it, as several defenses are always necessary.
Updating identity documents that have been mostly unchanged since the 15th century should be a priority for governments, joked Gorohhov.
The discussion of real-world attack vectors and mitigation methods became quite detailed, and panelists expressed optimism that effective countermeasures for sophisticated attacks are known, although they also cautioned against underestimating attackers or not anticipating the maturation of their methods.
The second session of the day focused on the perspective of users of government ecosystems of cybersecurity, telecommunications and financial services.
A session on testing and auditing followed.
NIST biometrics evaluator Patrick J. Grother spoke about the current state of the art in facial biometrics and risk mitigation. The latter includes creating prompts for users that humans can understand, but automated systems cannot, to prevent the possibility of parodies interpreting and following instructions correctly.
Kevin Carta of the French biometrics laboratory CLR Labs reviewed the threat of attacks by injection of biometric data, prepared or real. The injection is possible because the current architectures do not make it possible to associate images with a particular identified camera.
Biometrics must therefore be deployed against injection attacks. PAD systems, however, are not designed to recognize this type of attack. According to Carta, specific methods for detecting attacks by injection of biometric data must be developed to counter future developments in the type of fraud.
An international standard is being developed, he says.
Clemens Wanko of TÜV TRUST IT GmbH presented the auditor perspective, including how identity service providers are audited for compliance with international standards.
Clear benchmarks are needed to apply specifications at different levels of assurance to move the field forward, he says. Changes to eIDAS did not help with clarity.
Certicar.es Technical Director Paloma Llaneza delved into the complexities of overlapping standards and regulations, each of which must be regularly updated.
Back to the technical specification ETSI TS 119 461 for electronic signatures and infrastructures for trust service components providing proof of identity.
Hugo Mania from ANSSI gave an overview of the certification system and its objectives, and Dr. Christian Berghoff from German BSI described the biometric authentication component of the certification.
“AI systems have complex supply chains and are quite sensitive to small changes, which means there are different ways to attack them,” Berghoff warns.
He advocates manual inspection of at least some samples and measures to prevent full automation of attacks.
Sylvie Lacroix from Sealed explained how technical standards, certifications and regulations fit together for digital identity verification providers, and Jon Ølnes from Signicat discussed the scope of standards and regulations in areas beyond beyond trust services, such as their impact on financial service providers wishing to integrate users in a neighboring country.
According to Ølnes, knowing which rules apply, and even if they exist, remains a challenge for many service providers trying to transact across European borders.
A more unified set of requirements that still protects people and businesses from fraud is certainly possible, based on the tools and expertise discussed at the event. For now, it’s a work in progress.
biometrics | CLR Laboratories | cybersecurity | fake fake | digital identification | eIDAS | facial biometrics | fraud prevention | identity document | identity verification | IDnow | NIST | standards | To verify