If AI changes identity fraud every month, how does the industry know what actually works?

Imagine this scenario.

In January, an identity verification solution successfully completed an independent evaluation against the industry’s latest standards. The vendor announces the achievement, customers gain confidence and procurement teams add another checkmark to their selection process.

In February, someone discovered a new AI-generated document attack.

In March, another organization identified a novel biometric injection attack.

In April, a new open-source AI model dramatically improved the quality of deepfakes.

By June, fraudsters have automated many of these techniques.

Yet organizations are still relying on the confidence they gained from an evaluation completed in January.

Nothing is wrong with that evaluation. It accurately measured how the system performed against the threats that existed at the time. The challenge is that AI is compressing the time between new attack techniques, while most testing and certification programs still operate on much longer cycles.

This raises a broader question for governments, enterprises, technology vendors and independent laboratories: How should we evaluate identity systems when the threat evolves faster than the evaluation itself?

The evidence is difficult to ignore. Generative AI is making it easier to create convincing identity documents, faces, voices and videos. Publicly available models continue to improve while becoming accessible to a broader audience. Massive data breaches provide criminals with the information needed to create synthetic identities and execute account takeover attacks. Meanwhile, AI-assisted phishing, social engineering and device compromise continue expanding the attack surface.

The challenge is no longer simply building stronger identity systems. It is ensuring we can continuously measure whether they remain effective.

For years, the identity industry has relied on independent evaluations conducted by organizations such as NIST, DHS and accredited laboratories to benchmark technologies and build trust. These programs remain indispensable. They provide objective measurements, encourage innovation and help buyers make informed decisions.

However, the future of identity assurance cannot rely solely on static snapshots. It requires a continuous learning model.

The first step is recognizing that evaluation datasets must evolve continuously. Every significant advancement in AI introduces new attack techniques that should eventually become part of future evaluations. Yesterday’s dataset becomes tomorrow’s baseline.

Second, no single organization can realistically build and maintain comprehensive evaluation datasets on its own. Creating diverse, representative and current attack datasets requires significant expertise, investment and ongoing maintenance. This presents an opportunity for governments, enterprises, vendors, academia and independent organizations to collaborate on shared evaluation assets that strengthen the entire ecosystem.

More importantly, the industry must become better at converting newly discovered attacks into shared knowledge. It should matter less who identifies an emerging attack and more how quickly the broader community can responsibly learn from it. Whether a new technique is discovered by a government agency, an enterprise, a vendor, an independent evaluation partner or a researcher, the goal should be the same: update evaluation methodologies, measure resilience, identify gaps, improve defenses and share those learnings with the ecosystem.

Perhaps it is also time to rethink the role of independent laboratories. Rather than viewing them simply as testing labs that issue point-in-time certifications, governments, enterprises and vendors should increasingly engage them as evaluation partners. Their role extends beyond validating compliance. They can help continuously assess emerging threats, refresh evaluation datasets, measure progress over time and provide independent evidence that technologies are improving.

This shift becomes even more important as the industry raises the bar. Capabilities that once differentiated solutions are increasingly becoming baseline expectations, while evaluation partners are expanding their ability to assess higher-assurance presentation attack detection, biometric injection attacks and other emerging threats.

The future of identity assurance is not a single certification. It is a continuous cycle of evaluating, measuring, identifying gaps, improving and repeating.

Autonomous AI agents will only accelerate the pace at which new attack techniques emerge. The organizations that build the greatest trust will not be those with the oldest certification hanging on the wall. They will be those that learn, adapt and demonstrate resilience faster than the threats themselves evolve.

Article Topics

AI fraud  |  biometric testing  |  biometrics  |  digital identity  |  identity assurance