Ilayer research
A multi-agent pipeline that finds real vulnerabilities
We built an autonomous, multi-agent system that audits smart contracts and surfaces real high and medium findings on live protocols. It runs several models against each other and filters every candidate hard before a human sees it, so the signal is high and the noise is gone.
Sector: AI for security
Where this comes from
The studio's own security research pipeline, not a client deployment. Gates filter candidates; they do not prove a finding is exploitable.
Where this comes from
Same pipeline. Adversarial verification lowers the false-positive rate; it cannot raise the true-positive rate on bugs nobody looked for.
Where this comes from
Protocol names are withheld by choice. The severity labels are the platforms' own, not ours.
The situation
- Automated audit tools drown you in false positives. A finding is only useful if it survives scrutiny.
- One model is not enough. Different models miss different bugs.
- A candidate has to be filtered hard before a human sees it, or the tool wastes more time than it saves.
What we did
- 01A multi-agent hunt runs several models over the same target and cross-checks their findings against each other.
- 02A seven-gate pre-submission stack every candidate has to clear, plus a feedback loop that learns from what gets accepted.
- 03Adversarial verification: each finding is attacked by independent agents whose job is to refute it, and only survivors ship.
What held
- The same adversarial, multi-agent pattern is how we verify claims and evals across every engagement.
- Findings are filtered by refutation, not volume, so a reviewer's time goes to real bugs.
- Accepted findings feed back into what the pipeline hunts for next.
Start a project
Name the workflow that eats your team's week
Name the workflow that eats your team's time. We will tell you honestly whether AI fits, and if it does, scope a first piece with a known price.