Skip to content
AS
09 / AI & Data2026

AI-Augmented Binary & Source Triage

A multi-agent pipeline for decompilation review, high-risk source analysis, CWE mapping, and exploitability scoring.

Local LLMsAgentsGhidraCWE
Case file
Role
Solo build — research, engineering, documentation, release
Status
v0.2.0 · active
Year
2026
Position
09 of 13
Case 09 / ai-vuln-auditorArchitecture · Evidence · Signal

01 — Context

The problem worth solving.

Manual reverse-engineering review is expensive; automation must narrow the search without pretending a model is proof.

02 — Approach

How it was built.

  1. 01

    Partition binary and source review by specialised analysis roles.

  2. 02

    Identify memory-corruption, command-injection, and access-control candidates.

  3. 03

    Generate structured findings and harness templates for manual verification.

03 — Architecture

The shape of the system.

01Input

Decompiled functions, binaries, and high-risk source modules.

02Agents

Specialists for sinks, control flow, reachability, and exploitability.

03Review

CWE / ATT&CK mapping with human verification before publication.

04 — Evidence

Receipts, not adjectives.

CWEfinding taxonomy
Agentsanalysis roles
PoCverification output

05 — Hiring signal

“Uses agentic system design to accelerate research while preserving an evidence gate.”

Ships working systemsDocuments the methodExplains the trade-offsKeeps evidence attached

06 — What's next

Still being built.

Add benchmark corpora and measure precision on known vulnerable functions.