Key takeaways

• Zero knowledge proofs enhance transparency while maintaining confidentiality in decision-making processes.
• Lagrange is pioneering the integration of zero knowledge proofs in AI to ensure privacy.
• Current privacy methods in AI fall short compared to innovative cryptographic approaches.
• Privacy-preserving models require cryptographic integration from the development stage.
• Open source models often underperform in commercial applications.
• Protecting both intellectual property and consumer data is crucial in AI model deployment.
• Many private AI solutions fail to enhance privacy on commercially relevant models.
• The tech industry’s focus on trivial applications detracts from addressing national security.
• ZK machine learning relies on mathematics, not hardware, for privacy.
• Venture financing in aerospace and defense is insufficient for national security needs.
• Lagrange’s Deep Proof library is a key innovation in safeguarding AI data.
• Zero knowledge technology is reshaping the landscape of cryptographic security.

Guest intro

Ismael Hishon-Rezaizadeh is CEO and co-founder of Lagrange Labs, where he leads the development of DeepProve, the world’s fastest zkML library for verifiable AI inference. He spearheaded the first zero-knowledge proof of Google’s Gemma3 large language model, demonstrating production-ready verification at 158x the performance of competing solutions. His work advances ZK technology from crypto applications to defense-critical uses like securing autonomous drone swarms.

The role of zero knowledge proofs in decision-making

• Zero knowledge proofs provide transparency in decision-making while ensuring confidentiality.

• The zero knowledge proofs provide this way to glimpse inside the black box determine that the output that you got back has come from the correct model with these set of correct inputs

  — Ismael Hishon-Rezaizadeh

• These proofs allow for understanding the circumstances under which decisions are made.
• They help in adjusting decisions based on verified inputs and outputs.
• Zero knowledge proofs are crucial for secure cryptographic applications.
• They enhance trust in automated decision-making systems.

• Zero knowledge proofs allow for transparency in decision-making processes while maintaining confidentiality.

  — Ismael Hishon-Rezaizadeh

• Understanding their technical aspects is essential for implementing secure systems.

Lagrange’s innovative approach to AI privacy

• Lagrange is at the forefront of integrating zero knowledge proofs into AI.

• Lagrange likes to think of itself as the preeminent company for frontier research in applied cryptography.

  — Ismael Hishon-Rezaizadeh

• The company focuses on first principles innovation in cryptography.
• Lagrange’s Deep Proof is a zero knowledge machine learning library.
• This approach ensures privacy in AI without relying on air-gapped systems.

• We are uniquely positioned to build zero knowledge proofs.

  — Ismael Hishon-Rezaizadeh

• Their methods surpass current privacy solutions in AI.
• Lagrange’s innovations are crucial for the future of AI privacy.

The limitations of current AI privacy solutions

• Existing methods for AI privacy are inadequate compared to cryptographic innovations.

• A lot of the incumbent attempts to add privacy to AI are based on air gapped systems.

  — Ismael Hishon-Rezaizadeh

• These methods do not rely on first principles innovation.
• Privacy-preserving models require cryptographic security from the development stage.

• If you wanna use frontier closed source models you have to be able to bake cryptographic security in at the level of the model developer.

  — Ismael Hishon-Rezaizadeh

• Current solutions often fail to protect commercially relevant models.
• Many rely on open source models that underperform in commercial settings.
• There is a need for better privacy integration in AI models.

The need for cryptographic security in AI models

• Privacy-preserving models require cryptographic integration from the start.

• Zero knowledge machine learning does not rely on specific hardware for privacy.

  — Ismael Hishon-Rezaizadeh

• Open source models are often unsuitable for commercial applications.

• The performance of existing open source model is not good.

  — Ismael Hishon-Rezaizadeh

• Proprietary models need built-in cryptographic security.
• Protecting intellectual property and consumer data is crucial.
• Many private AI solutions fail to enhance privacy on proprietary models.
• Cryptographic security is essential for effective privacy preservation.

The dual need for privacy in AI

• AI models require privacy for both intellectual property and consumer data.

• You’re not gonna open source the best AI model.

  — Ismael Hishon-Rezaizadeh

• Protecting intellectual property ensures competitive advantage.
• Consumer data privacy is crucial for trust and compliance.
• Many private AI solutions rely on open-source models.

• A lot of the private AI solutions are just taking open source models that are publicly available.

  — Ismael Hishon-Rezaizadeh

• These solutions often do not enhance privacy on commercially relevant models.
• Effective privacy measures are needed for proprietary AI models.

The tech industry’s focus and national security

• The tech industry often prioritizes trivial applications over national security.

• A generation of businesses pursuing incremental and trivial applications of technology.

  — Ismael Hishon-Rezaizadeh

• This focus detracts from addressing critical national security issues.
• There is a need for tech innovation that materially influences defense.
• The current state of venture financing in traditional sectors is inadequate.

• There was very little venture financing that deployed into anything in traditional sectors.

  — Ismael Hishon-Rezaizadeh

• This gap affects national security and technological advancement.
• More investment is needed in aerospace and defense sectors.

The mathematical nature of ZK machine learning

• ZK machine learning is fundamentally mathematical, not hardware-dependent.

• ZK machine learning is entirely just mathematics.

  — Ismael Hishon-Rezaizadeh

• This independence from hardware is crucial for privacy applications.
• Mathematical approaches ensure provable AI privacy.
• Hardware constraints do not limit ZK machine learning.
• This approach is essential for scalable privacy solutions.
• ZK machine learning offers robust privacy without specific hardware.
• Its mathematical nature is key to its potential applications.

The implications of venture financing on national security

• Venture financing in traditional sectors is insufficient for national security.

• That’s not a successful state of a country.

  — Ismael Hishon-Rezaizadeh

• This lack of funding affects technological innovation in defense.
• More investment is needed in aerospace and defense industries.
• The current state of financing hinders national security advancements.
• There is a need for strategic investment in critical sectors.
• Addressing this gap is crucial for future security and innovation.
• National security requires a robust venture financing ecosystem.