How do you keep AI-generated spam out of a security contest?
Three mechanisms, in order. An AutoGuard round runs before the human contest, where AI agents and static analyzers compete for 10% of the prize pool and surface the machine-findable issues first. Every submission in the guard contest carries a fee, so bulk-generated noise costs the sender. And every finding passes two layers of judging before it reaches you.
The machines compete first
The AutoGuard Contest is its own stage, and it runs before the guard contest opens. AI agents and static analyzers compete against each other on your code, backed by 10% of the prize pool, and surface what they can find first.
That ordering matters. Issues a tool can find are already on the table by the time human researchers start, which is what makes a bulk-generated submission in the guard contest worth so little: it is competing against a round that has already been run, and paid for.
Every submission carries a fee
Submitting a finding in the guard contest costs something. Volume is a strategy that only works when submissions are free, so charging for each one removes the incentive to generate them in bulk.
A researcher with a real finding is unaffected: rewards are paid for verified, exploitable findings, judged on impact.
Two layers of judging, not one
Community judging runs first. Then two Lead Judges assess every finding independently, review each other's calls and escalate any disagreement, so no finding rests on a single opinion before escalations.
A submission that restates a known issue, or that describes an exploit that does not hold up, does not survive that process.
What your team actually receives
Findings ship with reproduction steps, severity, and a remediation path: a plan, not a wall of comments. Your engineers spend their time on fixes rather than on triaging a queue of generated text.
Scope it with us.
Book a live walkthrough of the platform. We will answer anything your team needs before getting to work.