David Strnadel

Security researcher · Czech Republic

I find security flaws in Apple software, so they get fixed before anyone can misuse them.

David Strnadel, smiling, arms crossed, in a garden in the Czech Republic

In plain words

What does a security researcher do?

Every app on your phone is supposed to follow rules, such as “this app must not read that data”. I look for places where such a rule quietly fails.

  1. Find

    I compare versions of iOS and macOS and look for a security check that is missing.

  2. Prove

    I build a small test app that shows the problem on a real device, with evidence instead of guesses.

  3. Report

    I send the evidence to Apple privately. Apple fixes the flaw for everyone and publicly credits the finding.

Selected investigations

Small details. Real consequences.

Two investigations, three credited vulnerabilities. Each starts with a plain summary; the technical detail is one click away.

All write-ups
iOS kernelCVE-2026-65398

A forgotten safety check in the core of iOS

According to Apple, an app could use it to crash the device or corrupt kernel memory, the most protected part of the system.

While comparing iOS builds, I noticed that a new operation in IOMobileFrameBuffer lacked a condition present in a related one. That difference became an out-of-bounds hypothesis, a proof of concept, and a report to Apple.

Technical summary

From difference to hypothesis

I used binary diffing to narrow down security-relevant changes during the iOS beta cycle. Comparing related operations helped isolate a missing condition and formulate a testable out-of-bounds hypothesis.

Test, then report

I developed an on-device proof of concept and documented the result for Apple. I directed the investigation and used AI to accelerate reverse analysis, PoC development, and the technical report.

Public outcome

Apple credited the issue as CVE-2026-65398, with shared attribution. Its advisory describes an out-of-bounds access issue that could allow an app to cause a system crash or corrupt kernel memory.

Privacy · SandboxOne report · Two CVEs

An app read data it should never have seen

An app locked in its own sandbox could still read data from other parts of the system, including information that could help identify the user.

The sandbox denied direct file reads, while public preference APIs in AuthKit and Accessibility returned data from other domains.

Technical summary

Compare the access paths

I built an on-device proof of concept to compare denied direct file reads with service-mediated preference access. A separate WidgetKit extension provided an independent process for checking the behaviour.

Rule out the wrong conclusion

Cross-process controls helped distinguish data exposure from writes that remained in a process-local cache. Those writes did not establish a cross-process write capability.

One investigation, two issues

Apple split the original report and tracked the resulting issues under AuthKit (CVE-2026-43730) and Accessibility (CVE-2026-43664). The Accessibility credit is shared with other researchers.

Credited by Apple

Each credit can be verified in Apple’s own security advisories.

CVE-2026-65398IOMobileFrameBufferOut-of-bounds access · Shared credit
CVE-2026-43664AccessibilitySensitive user data · Shared credit
CVE-2026-43730AuthKitPermissions issue · User fingerprinting

AI research

Ideas are a start. Experiments decide.

Alongside security, I study how AI models learn, and publish the results honestly, including when an idea doesn’t work.

Preprint · 2025

VQE-in-GAN

Can a quantum-inspired idea help generative AI models train better?

I explored VQE-inspired auxiliary losses in GAN training using PyTorch and Qiskit simulation. Controlled comparisons found no measurable benefit over simpler classical alternatives in the studied setting.

Simulator-based study. Published methods, ablations, and a negative result.

Ongoing · Master’s research

JEPA × spiking networks

Could brain-inspired computing benefit from a modern way of learning?

I investigate whether JEPA-style learning could benefit neuromorphic computation, comparing conventional and spiking neural networks through software experiments.

PyTorch experiments and Lava simulation. No experiments on physical Intel Loihi hardware.

About me

Curiosity, with controls.

I’m David, a security researcher from the Czech Republic. For five years I tested embedded software for the automotive industry at ZF Group. Hunting for the case nobody had tested turned out to be the best training for security research.

Today I study for a master’s in cybersecurity at Tomas Bata University in Zlín and research Apple platforms independently. I like taking a system apart, finding the assumption that matters, and building an experiment around it.

AI helps me move faster. Reproducible evidence is how I decide whether an idea holds up.

2026 — presentIndependent security research
Apple platforms, reverse engineering, and AI-assisted vulnerability discovery.
2026Revelia
AI Analyst / Data Scientist / Project Manager.
2021 — 2026ZF Group
Software testing in embedded and AUTOSAR environments.
EducationTomas Bata University in Zlín
BSc in Software Engineering, 2025.
MSc in Cybersecurity.

Let’s talk.

Building on Apple platforms, running a security team, or researching similar problems? I’m happy to talk about vulnerability research, reverse engineering and turning a hunch into reproducible evidence.