field notes · 202 pages · 23 labs
By the last page you'll be able to spot the representation gap that drains a bridge — rebuild the exploit in Foundry, watch the invariant break, and patch it — and you'll know, with a number, exactly where an LLM stops being able to do it for you.
In February 2022 the Wormhole bridge lost $320,000,000 to sixty-five lines of Rust that worked perfectly — in a world that didn't exist. This book rebuilds ten real exploits (~$1.2B in losses) and reports what the tools catch.
Secure checkout via Gumroad · refundable if it's not for you
6×9 typeset PDF — designed for screen and print
They're representation gaps: the space between what the code believes and what the chain can prove. Wormhole checked a signature on one account type and forgot the other. The contract worked. The world it assumed didn't exist.
I wanted to know if a language model could see that gap — not in the abstract. I wanted a number. So I built a benchmark: ten real bridge exploits, six classes of compositional vulnerability, a Foundry suite that reproduces each one from source. Then I ran the tools.
Static analyzers: ≈0%. An LLM: 40%. This book is the investigation that produced those numbers — and a mechanistic answer to what lives in the other 60%.
Sixty-five lines of Rust. The contract compiled, passed its tests, and ran in production for months. It checked a signature on one account type and never checked it on another — and for a while, nobody needed it to. The bug wasn't in the code. It was in the gap between what the code believed and what the chain could actually prove.
That gap has a price. On February 2nd, 2022, it was $320 million. This book is about learning to see it before the attacker does — and about whether a machine can be taught to see it too.
202 pages · 23 chapters · written as field notes, ordered the way the investigation actually unfolded.
Real exploit anatomy, rebuilt and proven in Foundry: the honeypot with a freeze button, the bridge that paid twice, the bridge that minted from nothing. Plus the EIP-712 bug I found and fixed — the PR is merged in alloy-rs/core.
A dice game with no win condition. Why a coin flip can be verifiable and still not trustless. What a ZK proof actually proves, ZK dark chess with real fog of war, FHE where ZK stops — and the post-quantum scheme Shor breaks anyway.
The benchmark itself: who audits the auditor, what active learning buys you, a model that asks when it can't explain what it sees — and the interpretability hunt for why the model reads "not" but can't use it.
+ Preface · and a closing chapter, “The Gap”
Clone the repo, run the exploit, watch the invariant go red — then fix it and watch it go green. Real Foundry suites against real bug classes.
$ forge test --match-contract WormholeReplay
[FAIL] test_invariant_mintRequiresProof() ✗ minted 120,000 wETH from nothing
$ git apply fix/check-sysvar-account.patch
$ forge test --match-contract WormholeReplay
[PASS] test_invariant_mintRequiresProof() ✓ invariant holdsSolidity / DeFi developers who want to internalize how real exploits compose — not CTF puzzles, the actual mechanisms behind nine-figure losses.
Auditors who want a falsifiable benchmark for what LLM tooling can and cannot catch today.
ML engineers pointed at security problems, who want to know where the model's understanding actually stops — measured, not vibes.
Anyone who's read one too many "AI will replace auditors" takes and wants the number instead.
Not a Solidity tutorial (you should read code), not a course (Cyfrin Updraft is free and excellent), and not a prediction about AGI. It's field notes from pointing the current generation of models at the hardest bugs in the most adversarial environment we have — and a precise account of where they fail.
I'm Toma. I didn't write a book about exploits I read about — I rebuilt ten of them from source in Foundry until every invariant went red, then went looking for what the tools miss. Along the way I found a real EIP-712 encoding bug in alloy-rs/core and fixed it.
The PR is merged. That's the standard this book is held to: if a claim isn't reproducible, it isn't in here.
A 202-page typeset PDF (6×9, built for both screen and print), plus links to every lab repo and the benchmark suite. Instant download after checkout.
You should be comfortable reading code. It's not a from-scratch Solidity tutorial — if you've never touched a smart contract, start with Cyfrin Updraft (free) first, then come back. The labs assume you can run forge test.
Real — ten historical bridge exploits totalling roughly $1.2B in losses, reproduced from source in Foundry. The benchmark numbers (static analyzers ≈0%, LLM 40%) come from running actual tools against those reproductions.
Yes — free updates to this edition land in your Gumroad library automatically. Refunds are handled through Gumroad if it's not what you expected.
Instant download · secure checkout via Gumroad · refundable if it's not for you