Research

Watch · narrated whiteboard episodesL3

Below the Radar: Attacks That Break the Assumptions Nobody Names

Most security education recycles the same dozen attacks. This threat lab covers the ones that break an assumption you never knew you were making — starting with the deepest of all: that two systems agree on what the same bytes mean. When a guardrail, a language model, and a tool each parse one input differently, the exploit lives in the disagreement, not in any single component. Each class is taught by the assumption it violates, the mechanism that makes it work, and the assumption-free defense — grounded in the seminal paper that named it, and tied back to the AI-agent stack every time.

Murali Chillakuru·10 episodes
  1. 18 min Episode 1Parser Differentials: When Two Systems Disagree About What the Bytes MeanA whiteboard conversation on the attacks that live in the gap between two components that read the same bytes differently — and why a modern A I agent is that gap by construction.
  2. 18 min Episode 2Time Is a Weapon: TOCTOU and the Gap Between Check and UseA whiteboard walkthrough of the time-of-check to time-of-use race — from a 1996 file-system attack to an AI agent that acts on an approval that already expired.
  3. 17 min Episode 3When Physics Leaks Secrets: Microarchitectural and Timing Side ChannelsA whiteboard walkthrough of how timing, caches, and speculation leak secrets your code never returned — from Kocher's 1996 stopwatch to a shared inference GPU.
  4. 18 min Episode 4Bit-Flips and Fault Injection: From Rowhammer to Model-Weight CorruptionA whiteboard walkthrough of how a stored bit can flip with no write at all — from Rowhammer escalating to root to a handful of flips backdooring a model.
  5. 18 min Episode 5Covert Channels and the Air-Gap Myth: Exfiltration Without a WireA whiteboard walkthrough of how secrets escape a disconnected machine through physics nobody modeled — from Lampson's confinement problem to an emanation catalog and its defenses.
  6. 17 min Episode 6Trusting Trust: Compiler and Build-Chain BackdoorsA whiteboard walkthrough of Ken Thompson's self-propagating compiler backdoor and its modern descendants — and how reproducible builds and provenance claw the trust back.
  7. 18 min Episode 7The Myth of Deletion: Data Remanence and Model MemorizationA whiteboard walkthrough of why 'delete' rarely destroys — from cold-boot key recovery to a model that memorized your data and cannot unlearn it — and how to design so forgetting is real.
  8. 18 min Episode 8Subliminal Channels: Instructions Hiding in Plain SightA whiteboard walkthrough of hidden messages riding inside perfectly valid ones — from Simmons' prisoners and kleptographic keys to invisible instructions that hijack an AI agent.
  9. 18 min Episode 9Economic and Incentive Attacks: Breaking the Rules of the GameA whiteboard walkthrough of attacks that break no rule at all — from Goodhart's law and MEV to reward hacking — where an optimizer perfects the metric while the true goal quietly dies.
  10. 18 min Episode 10The Hidden-Assumption Threat Register: Auditing Your Own StackA whiteboard synthesis of the whole series — turning nine classes of below-the-radar attack into one reusable register of named assumptions, observable tells, and assumption-free controls.