Ariel Koren Get in Touch
The Work

A curated archive of security research, systems, and product work.

Mixed entry types - disclosures, CVEs, architecture, and founder work - organised, not inflated.

12 entries 2016-2026 Updated Jul 2026

Security Research

Public disclosures, CVEs, coordinated disclosure, and AI / agentic security work.

3entries · 2020-2026
2026
Vulnerability research Public / Vendor-declined

Skia RenameFont OOB. Google Won't Fix It.

A manual bounds-audit of Skia's SkOTUtils::RenameFont found an attacker-controlled heap buffer overflow write from a crafted font, reachable through the public SkFontMgr_New_GDI() API on the Windows GDI backend. Verified end to end under ASan (243,300-byte write into a 280-byte buffer) with a one-hunk fix. Google declined it won't-fix: the crash was not disputed, but a security fix requires web-reachability and the web path (DirectWrite) never calls RenameFont. No CVE; unpatched in main.

Read the writeup
2026
Vulnerability research Public / Patched

CVE-2026-48029: Two Grid-Decode Vulnerabilities in libheif

An afternoon of fuzzing libheif 1.21.2 produced two memory-safety bugs in the same grid-decode function: a NULL pointer dereference that is a deterministic DoS from any public decode API, and a uint32 underflow that becomes a heap out-of-bounds read in release builds. Reported privately; fixed in libheif 1.22.0 within three weeks.

Read the writeup
Systems Design & Writing

Security architecture, fuzzing, things I built - and essays on AI and how engineering work is changing.

3entries · 2026
2026
Engineering leadership Essay AI

AI Slop Is a Leadership Problem

AI collapsed the time to a working solution, but the time to real understanding did not move. An essay arguing that slop comes from lack of direction, not from AI - when execution gets cheap, leverage shifts to whoever defines the problem and holds the quality bar.

Read the essay
2026
Research platform architecture Writeup AI

Anvil: An Autonomous Vulnerability-Research Platform

Not an AI that finds bugs - a research platform where AI agents do the repetitive glue work and deterministic infrastructure decides what is true, including an evidence ladder that refuses to call a sanitizer crash a vulnerability. The architecture, a libpng walkthrough, and an honest read on the findings so far, including CVE-2026-48029.

Read the writeup
2026
Agent security architecture Essay AI

LLMs Are Probabilistic. Agent Authority Cannot Be.

Agents now take real, privileged, sometimes irreversible actions, but the models driving them only ever predict. A design argument for compiling the user's request into a scoped task policy before untrusted context arrives, then enforcing it outside the model.

Read the writeup
Startups

SNDBOX, OmniBoard, and founder / product work - including the post-mortems.

1entry · 2023-2024
Archive

Older technical writeups - malware research, RE notes, and short technical pieces.

5entries · 2016-2020
2017
Reverse engineering Writeup

Fusion - Level 06 solution

Walkthrough and exploit for Level 06 of the Fusion exploitation series. Race conditions, integer overflow, stack overrun.

Read
2016
Malware research Writeup

Ursnif malware: deep technical dive

Reverse-engineering walkthrough of Ursnif: process injection, sandbox evasion, traffic obfuscation, and a bug in the malware's own DGA.

Read