Skip to content
PL
Portrait of Karol Budzyn

Cryptography · Quantum-safe networks · AI security

Karol Budzyn

Principal R&D Architect · Orange Polska · Warsaw, Poland

I am a cryptography and security researcher based in Warsaw. As Principal R&D Architect at Orange Polska, I work on the long-term security of telecommunication networks: the migration to post-quantum cryptography, the integration of quantum key distribution with classical key management, and the security of AI systems that act autonomously.

My work sits between standards and deployment — how new primitives such as ML‑KEM and ML‑DSA behave inside real protocols, how trust can be established between AI agents, and how large systems stay cryptographically agile. This site collects my papers, research projects and longer-form writing.

Email arXiv GitHub LinkedIn

Research

B LLL(B)
Lattice basis reduction (LLL): same lattice, shorter basis. Lattice-based schemes rely on how hard this gets in high dimension.

Post-quantum cryptography

Migration of network protocols to the NIST standards (FIPS 203–205), hybrid key establishment, KEMTLS, and crypto-agility in long-lived infrastructure.

A B A B k + + 1 × × 0 + × · × × 1 + + 1 + × · × × 0 + + 1
BB84: photons sent in random bases; where Alice’s and Bob’s bases agree, the bits become the sifted key.

Quantum key distribution

Integrating QKD links with classical and post-quantum key management in telecom networks, trusted-node architectures, and the practical limits of quantum-secured transport.

verify ✕ a₁ a₂ a₃ a₄ E
Signed envelopes between agents: a message altered in transit fails verification and is dropped.

Security of AI systems

Identity, authenticated communication and verifiable behaviour for agentic AI; prompt injection and data exfiltration in multi-agent settings.

Selected publications

View all →

Recent writing

View all →

  1. Zero-Knowledge Proofs: How to Convince Someone Without Saying Anything

    ZKPs let you convince a verifier that a statement is true while revealing zero information beyond that fact. Not an approximation, a strict mathematical guarantee. Here is how it actually works.

    13 min read · Cryptography, ZKP, Zero-Knowledge

  2. Agentic AI and LLM Security: What Changes When the Model Can Act

    LLMs that browse the web, write code, and call APIs are a fundamentally different threat surface than chatbots. This post breaks down the attack classes unique to agentic systems and what defenders actually need to think about.

    11 min read · AI Security, LLM, Agentic AI

  3. Cyber Resilience in Banking: DORA, FFIEC, and SWIFT's Security Controls

    A practitioner's guide to the major cybersecurity frameworks governing financial institutions - DORA in the EU, the FFIEC guidelines in the US, and the SWIFT Customer Security Programme worldwide.

    8 min read · Banking Security, DORA, FFIEC

  4. PCI DSS 4.0: What the Standard Actually Demands

    A practitioner's breakdown of PCI DSS v4.0 - its twelve requirements, the 2025 deadline for future-dated controls, and where organisations most commonly fall short.

    7 min read · PCI DSS, Compliance, Payment Security