Risky Business Features

A Risky Biz Experiment: Hunting for iOS 0day with AI

Mar 31, 2026
Karina Klaur, an OpenClaw AI assistant acting as a technical thinking partner for vulnerability analysis. She inspects WebKit, hunts a suspected audio decode bug, then uncovers an SVG buffer overflow. Short, sharp scenes cover chaining primitives, sandbox and PAC hurdles, JIT corruption theory, refusal to produce exploit code, and a walkthrough of the multi‑stage Karuna chain.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Media Decoders Commonly Risk Size Mismatch Overflows

  • Audio decoding often sizes output buffers from estimated frame counts, leaving a gap when actual decoded frames exceed estimates.
  • Floating point multiplications and conversions across types can amplify rounding errors and enable writing past allocated PCM buffers.
INSIGHT

Patches Can Be Platform Specific Leaving Alternate Paths Vulnerable

  • A guard existed only on some platform paths, so a vulnerability can persist on alternate codec paths.
  • In that commit the non‑WebM path used an assertion that crashed but the WebM path had asymmetric frame counters leaving edge-case gaps.
ANECDOTE

Thought Experiment Linking Audio Overflow To SVG Exploit

  • James and Karina mapped an audio overflow primitive into a broader chain conceptually usable with other bugs.
  • They considered using audio PCM writes to corrupt objects then leak addresses for ASLR bypass before precise SVG writes.
Get the Snipd Podcast app to discover more snips from this episode
Get the app