Principal Applied Scientist · Amazon Web Services

Tancrède Lepoint.

Cryptography at Internet scale.
AI security across AWS.

I co-designed the cryptography behind two NIST post-quantum standards and built privacy systems deployed by Amazon, Apple, and Google.

View selected work Get in touch CV PDF · 2 pages
Tancrède Lepoint
Cryptographer. Builder. Curious by default.
Selected work

Research, standards, and deployed systems.

All publications
01 / AI SECURITY

AI security research and engineering at AWS.

I develop continuous security assessment, methods to test and debug agentic applications, and evaluations of AI models' cyber capabilities.

I research neurosymbolic methods to identify incomplete or incorrect security fixes and propose remediations that provably address the underlying bugs.

Selected publications in AI & automated reasoning
From a question to a complete answer. An under-specified question leaves date and territory free. Symbolic execution finds 17 outcomes; one targeted follow-up reduces them to two cited answers, checked against the rule engine (NeSy 2026).
  • The question is a region Under-specified questions to rule engines define regions, not points. An LLM maps the stated facts to a symbolic region; concolic execution covers every reachable outcome, and Z3 verifies the resulting cited partition. When it is too large, the system asks one targeted follow-up. NeSy 2026.
  • Delta debugging for LLM-integrated systems Uses semantic markers and delta debugging to isolate input segments responsible for an LLM application's failure. In an AWS case study, a 502-paragraph input was reduced to three failure-inducing segments. ICSE-SEIP 2026.
  • Automatic detection of data entry errors Uses LLMs to compare source documents with structured entries, flag inconsistencies, and route uncertain cases to human review. ICSE STATIC 2025.
  • Safe validation of pricing agreements A protocol for validating AWS pricing agreements against customer intentions. ICSE-SEIP 2025.
02 / POST-QUANTUM CRYPTOGRAPHY

Protecting Internet communications at global scale.

I co-designed CRYSTALS-Kyber and Dilithium, adopted as NIST's ML-KEM and ML-DSA standards. ML-KEM now underpins the dominant post-quantum key exchange across global web infrastructure; ML-DSA standardizes quantum-resistant digital signatures.

My contributions span cryptographic design, security proofs, and implementation, including Kyber, Dilithium, and the earlier BLISS signature scheme.

A little noise, a hard problem: explore the cryptography
The hard problem underneath. Learning With Errors: multiply a public matrix A by a secret vector s, add small noise e, publish the result b. Recovering s from (A, b) is believed hard, even for a quantum computer. ML-KEM and ML-DSA are built on lattice problems of this shape.
  • CRYSTALS-Kyber and CRYSTALS-Dilithium Lattice-based key encapsulation and digital signature schemes, standardized by NIST as ML-KEM and ML-DSA. Kyber/ML-KEM is deployed in TLS, private messaging, and remote access; ML-DSA provides standardized quantum-resistant signatures.
  • Improved security proofs in lattice-based cryptography Introduced the use of Rényi divergence as an alternative to statistical distance for lattice-based security reductions, leading to tighter parameters. Best Paper, Asiacrypt 2015.
03 / PRIVACY-ENHANCING TECHNOLOGIES

Privacy-preserving AI at Google and Amazon.

I co-designed secure aggregation later deployed in Google's differentially private federated learning. The protocol hides individual device updates and reduces per-device computation from linear to logarithmic in the number of participants.

At Amazon, I lead the design of differentially private synthetic data pipelines for AI training. I also contributed machine-verified differential privacy algorithms deployed in AWS Clean Rooms for collaborative analytics.

Explore the protocols & deployments
Private information retrieval. The client’s query travels encrypted; the server computes over its whole database under homomorphic encryption and returns one encrypted record, without ever learning which one (Usenix 2021).
About & approach

A bit about me.

I co-designed cryptographic schemes that became global standards and built privacy systems deployed by Amazon, Apple, and Google. At AWS, I lead AI security research and engineering with teams across all AWS AI services.

The tools change: security reductions in lattice cryptography, machine-checked proofs in differential privacy, symbolic verification in AI. The discipline stays the same: explicit threat models, formal guarantees, and honest claims about what is and isn’t proven.

My Ph.D. at École Normale Supérieure and the University of Luxembourg received the Gilles Kahn Dissertation Award in 2014. Read the thesis ↗

A small graph-theory detour: my Erdős number is 3, through Claire Mathieu and Eli Upfal.

A path through research & engineering

  1. 2025–presentAmazon Web Services · AI Security

    Principal Applied Scientist
    Continuous assessment, agent testing, provable remediation research, and synthetic training data.

  2. 2022–2025Amazon Web Services · Provable Security & Automation

    Principal Applied Scientist
    Built data protection tools and an automated reasoning service to map data flows; contributed verified privacy algorithms to AWS Clean Rooms.

  3. 2021–2022Apple

    Cryptographic Engineer
    Cryptographic implementations in corecrypto, CommonCrypto, and CryptoKit; homomorphic encryption for private queries.

  4. 2018–2021Google

    Senior Research Scientist
    Private lookups, anonymous authentication, secure aggregation, and exposure-notification analytics. Contributed to the shell-encryption library.

  5. 2016–2018SRI International

    Senior Computer Scientist
    Co-designed Kyber and Dilithium; led privacy-preserving computation research under DARPA Brandeis.

  6. 2011–2016CryptoExperts

    Junior Security Expert
    Co-designed BLISS; contributed homomorphic-encryption research and implementations to the EU-funded HEAT project.

Download my CV (PDF) ↓
Open source

A side project

fhe.rs is my fully homomorphic encryption library in pure Rust, implementing the BFV scheme.

Explore fhe.rs on GitHub ↗
Recently

News & milestones

Appointed IACR IT Manager, a board-appointed position.
Delta Debugging for LLM-integrated Systems, ICSE-SEIP 2026.

Professional service

IT Manager IACR 2026–present (board-appointed)
Director IACR 2018–2024
Co-editor Cryptology ePrint Archive 2016–2023
General Chair CRYPTO 2024
Program Chair WAHC 2020 • WAHC 2019
Program committees & past service
Real World Crypto 2027EuroS&P 2026ASIACRYPT 2025Real World Crypto 2024USENIX SECURITY 2024Real World Crypto 2023CT-RSA 2023PKC 2023USENIX SECURITY 2023CRYPTO 2023USENIX SECURITY 2022CT-RSA 2021IEEE S&P 2021USENIX SECURITY 2021CHES 2021CT-RSA 2020USENIX SECURITY 2020CHES 2020EUROCRYPT 2020CT-RSA 2019PKC 2019USENIX SECURITY 2019SSR 2019CT-RSA 2018COSADE 2018AFRICACRYPT 2018PQCRYPTO 2018CRYPTO 2018SCN 2018SSR 2018CHES 2018WAHC 2018CT-RSA 2017WAHC 2017EUROCRYPT 2017CHES 2017PQCRYPTO 2017CANS 2017INDOCRYPT 2017WAHC 2016ICITS 2016CHES 2016WAHC 2015SECICT 2015CANS 2015ACNS 2015ACISP 2015

Get in touch

Have a question about my work, an idea to explore together, or a role in mind? Feel free to email me.

Email me ↗LinkedIn ↗Download CV PDF · 2 pages