yBorg
yBorg was a four-week yAudit fellowship about AI-assisted security. Each fellow built their own AI-powered security tooling, then put it to work on real targets: a group bug bounty hunt and live audits.
Before touching live code, fellows calibrated and benchmarked their tools on shadow audits of past public contests (Beedle and Mento V3), where the known findings give a baseline to measure against.
Along the way, guest speakers from teams such as Trail of Bits, OpenZeppelin and Flint Security shared how they use AI agents in their own audits and bug bounty hunting. Their sessions are collected below.
Resources
Recordings of the sessions held during the fellowship.
Week 1
Sep 28 – Oct 2
Sep 28 · 59 min
Fellowship kickoff
yAudit team
The goal, the mentors and the four-week plan: build your own AI-assisted security tooling, then use it for bug bounty hunting and live audits.
Sep 29 · 15 min
Audit flow exercise
0xFlint · Flint Security
A worked audit of a simple lending protocol, from what the protocol is trying to do to where its collateral rules break.
Sep 30 · 1h 10m
The framework trap
fucory · Smithers
Which agent frameworks age out and which last. Build on the primitives, not the topology, and keep a human in the loop.
Oct 1 · 1h 3m
One audit, from first look to report
Andrew Anderson, Josep · Hob, OpenZeppelin
A full audit in one workspace: brief the agent, map the codebase, hunt in parallel, triage and prove a lead, then write the report.
- No recording
Oct 2
Session with Ben Samuels
Ben Samuels · Trail of Bits
Week 2
Oct 5 – Oct 9
Oct 6 · 48 min
Session with 0xfirefist
0xfirefist
Reading a protocol from its architecture diagram first, then looping agents over it and choosing the right model for each step.
Week 3
Oct 12 – Oct 16
Oct 12 · 1h 35m
Setting yourself up for success in AI-assisted bug bounty hunting
Riptide · GregoAI
The groundwork before an agent hunts a bounty: pin the scope, rebuild the deployed code, ship a PoC template and an out-of-scope cheat sheet, then pick a program on Sherlock.
Oct 13 · 1h 15m
Thinking like an attacker
Dravee
A tour of the attacker's search space: how to bound it, which invariants to enforce, where the money goes, and where to hunt for highs and crits.
Session notes ↗
Shadow audits
Past public contests, packaged with pinned sources and dependencies, used to calibrate and benchmark the fellows' tools.
Audits
Live audits carried out during the fellowship.
Lynxify LYNX LP Range Manager ↗
October 2026
UFG Factory and Treasuries ↗
October 2026
Cap Network v2
October 2026 · Report is private