> For the complete documentation index, see [llms.txt](https://dualguard.gitbook.io/dualguard/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://dualguard.gitbook.io/dualguard/security-reviews/security-reviews-for-protocol-teams/agentic-security-review-contests.md).

# Agentic Security Review Contests

Agentic Security Review Contests are short public review windows where guards use AI agents to analyze a fixed scope under clear rules, incentives, and judging.

At DualGuard, this phase comes **before** the traditional security review contest. It helps surface faster wins early, reduce avoidable noise later, and focus the longer contest on deeper logic and higher-value attack paths.

### What this review is for

An agentic contest is useful when you want an early adversarial pass on a release candidate before the broader traditional contest begins.

It is designed to:

* Catch straightforward vulnerabilities early
* Stress-test assumptions with many parallel workflows
* Improve the signal quality of the full review sequence
* Encourage security researchers to invest time on AI x Web3 Security

This phase does **not** replace the traditional contest. It is the first layer in a broader pre-launch review process.

### How it fits into the review flow

The standard flow is:

1. Freeze scope and share the release candidate
2. Run the agentic contest
3. Judge and validate submissions
4. Run the traditional security review contest
5. Complete the [Mitigation Review](/dualguard/security-reviews/security-reviews-for-protocol-teams/mitigation-review.md)

For the full end-to-end process, see [How Security Review Contests Work](/dualguard/security-reviews/security-reviews-for-protocol-teams/how-security-review-contests-work.md).

### What protocol teams should expect

Before the contest starts, your team provides the same essentials required for any scoped review:

* Final commit, branch, or tagged release candidate
* Setup instructions and testing guidance
* Deployment context, assumptions, and known limitations

DualGuard then opens a short contest window where guards review the pinned scope with AI-assisted workflows and manual validation.

Your team should stay available in the shared channel. Fast answers matter. They reduce ambiguity and help guards spend time on real attack surface.

### What makes this phase different

The main difference is the review style.

In an agentic contest, guards can use AI agents to accelerate code comprehension, hypothesis generation, coverage mapping, and issue triage. The expected outcome is more parallel exploration in less time.

That changes the value of the phase:

* More breadth early in the process
* Faster detection of obvious and semi-obvious issues
* Better preparation for the traditional contest

The standard for rewards does not change. Submissions still need to describe a real issue in scope and survive judging.

{% hint style="info" %}
AI assistance increases throughput. It does not lower the bar for validity. DualGuard rewards confirmed findings, not raw tool output.
{% endhint %}

### What you receive from it

The agentic contest produces judged security findings against the frozen scope.

That gives your team:

* Earlier visibility into reachable issues
* A cleaner starting point for the traditional contest
* More time to plan remediation before launch

Prize handling for this phase is covered in [Prize Pool Distribution](/dualguard/security-reviews/security-reviews-for-protocol-teams/prize-pool-distribution.md).

### When to choose it

Agentic contests are a strong fit for:

* Mainnet launches
* Major upgrades
* Codebases that benefit from a fast first-pass before deeper human review

If your goal is a complete DualGuard pre-launch review, this phase should be treated as the opening layer, not the final one.

For scheduling guidance, see [Security Review Contest Timeline](/dualguard/security-reviews/protocol-teams/security-review-contest-timeline.md).