Skip to content
AS
11 / AI & Data2026

Agentic Patch Differ

A pipeline that compares pre- and post-patch binaries, isolates silent security fixes, and ranks reachable functions for review.

GhidraBinDiffCFGARM64Agents
Case file
Role
Solo build — research, engineering, documentation, release
Status
Planned · frontier build
Year
2026
Position
11 of 13
Case 11 / agentic-patch-differArchitecture · Evidence · Signal

01 — Context

The problem worth solving.

Patch analysis is slow when researchers must manually translate assembly-level changes into root-cause and exploitability hypotheses.

02 — Approach

How it was built.

  1. 01

    Diff control-flow graphs and changed basic blocks.

  2. 02

    Use agents to explain likely security fixes and map untrusted input reachability.

  3. 03

    Draft trigger inputs for manual exploit verification.

03 — Architecture

The shape of the system.

01Diff

Headless Ghidra and BinDiff exports feed a normalised CFG model.

02Reasoning

Agents translate low-level changes into candidate root causes.

03Ranking

Reachability and input provenance prioritise review order.

04 — Evidence

Receipts, not adjectives.

CFGdiff model
ARM64target ISA
PoCdraft output

05 — Hiring signal

“Connects reverse engineering, program analysis, and agentic research into one measurable pipeline.”

Ships working systemsDocuments the methodExplains the trade-offsKeeps evidence attached

06 — What's next

Still being built.

Implement function matching and benchmark against public patched vulnerabilities.