At DevConnect’s House of ZK, Nexus Chief Scientist Jens Groth took part in a fireside chat titled “Verifying Intelligence” to explore how zero-knowledge proofs (ZKPs) are reshaping the future of trustworthy computing. From the early breakthroughs in zk-SNARKs to today’s research frontiers in verifiable AI, Jens offered both historical depth and forward-looking insight into the evolution of the field, and Nexus’s role in shaping what comes next.
From theory to application
Jens began by reflecting on the arc of ZK research over the past two decades. What started as a deeply theoretical discipline has matured into a foundation for real-world systems.
“I started when ZK was a very general concept — someone had to sit down and define the first constructions. The real shift came around 2010, with fundamental breakthroughs like GGPR, which laid the groundwork for zk-SNARKs.”
These breakthroughs, once academic, are now embedded in scalable infrastructure and protocols. But Jens emphasized that we’re still early in the application curve, and that continued innovation will require closer alignment between academia and engineering.
The Nexus vision: From verifiability to economic security
When asked what drew him to Nexus, Jens pointed to the company’s ambition to apply ZK beyond cryptographic elegance and to reimagine core systems of trust at internet scale.
“Nexus had a vision that immediately resonated: a society where everything can be verified. That starts with verifiable finance — proving that transactions happened, that computation was correct — but it doesn’t stop there.”
He highlighted one critical area where verifiability alone falls short: economic security. Even if a system is provably correct, it still needs to withstand financial manipulation or adversarial game-theoretic behavior.
“Verifiability is fantastic, but it’s not everything. You also need to think about what happens outside the circuit—things like confidentiality, economic incentives, and the real-world context in which these systems operate.”
Turning to AI, Jens offered a nuanced take on the prospects of verifying machine learning models — especially in sensitive domains like healthcare, defense, and finance.
“Training runs are expensive. Verifying the full process isn’t feasible today. But we’re seeing smart workarounds and verifying the model’s output behavior, or certifying the integrity of specific computations within the pipeline.”
The goal isn’t to add ZK to everything indiscriminately, but to focus verification where it matters most. In Jens's view, that means developing flexible frameworks that adapt to the specific needs of a use case, whether that’s model provenance, data integrity, or auditability.
Beyond math: The social layer of trust
Jens closed by identifying a critical, often overlooked barrier to ZK adoption: human understanding and institutional trust.
“We’re doing great research, great engineering, but there are many other aspects we need to get right. People outside the space need quick, intuitive reasons to believe this is the safe choice. That means better standards, better education, and stronger social mechanisms.”
To scale verifiability, in other words, we need more than just proofs. We need a shift in mindset. Systems must not only be secure; they must feel trustworthy, usable, and aligned with the values of the people they serve.
Closing thought
Verifiability is no longer a niche feature, it’s becoming a foundational expectation in finance, AI, and beyond. But as Jens makes clear, achieving that future requires more than technical innovation. It demands a new contract between systems and society and one where correctness, accountability, and economic resilience are baked in from the start.
Daniel McGlynn
