Skip to main content
Devin Security finds vulnerabilities and helps you ship fixes. Open Security in the app at /security to start. From there you can start scans, review vulnerabilities on the Findings page and assign fixes to Devin, and track specific risks on the Threats page.

Access and permissions

Your organization must have access to security scans. Your role also controls whether you can start scans, manage scans, or manage findings. If Security is not available, ask an administrator to check your access.

How a scan works

When Devin scans a repository for vulnerabilities, the scan runs these steps:
  1. Threat model. Devin reads your repositories and works out what an attacker would target. See Threat modeling.
  2. Investigate. Devin searches the code for vulnerabilities, using child sessions to cover the codebase in parallel.
  3. Validate. Devin checks each candidate and discards the ones it cannot support with evidence.
  4. Report. Devin publishes the findings and ranks them against the threat model.
Scans that import findings from another tool skip the first two steps. See ingestion scans.

Start a scan

  1. Open Security and select Start a scan.
  2. Under Target repository, select the repositories to scan.
  3. Select Normal scan or Deep scan. See Scan modes.
  4. Optional: open Advanced to limit the scan to certain directories or file patterns, or to set up a recurring scan.
  5. Select Continue with Devin.
Devin opens a session and starts threat modeling. The scan starts after you approve the setup in that session. For scans that start on a schedule or on an event, see Automations.

Bulk scan

To cover many repositories at once, select Bulk scan in the scan form. Search for repositories, then select the ones to scan. Each repository gets its own scan, and Devin does not analyze the repositories together.

Scan modes

The scan mode sets how deeply Devin investigates your code.

Normal scan

A Normal scan balances speed and depth. It is the default, and it is the best choice for most repositories and for recurring scans. Limitations:
  • Devin spends less reasoning effort on each part of the code. A Normal scan can miss vulnerabilities that need several steps to exploit or that cross many files.
  • Devin validates findings in batches, so each finding gets less individual attention than in a Deep scan.

Deep scan

A Deep scan runs on frontier cyber models, including Claude Mythos. These models are built to find and reason about vulnerabilities. Devin reasons longer about each part of the code and validates each finding on its own. A Deep scan takes more time and uses more ACUs than a Normal scan. Your account team must turn on Deep scans for your organization. Until then, the Deep scan card is locked. Ask your account team for access. Limitations:
  • No custom instructions. You choose the repositories and the scope, but the text you write does not reach the scanner. This includes the answers you give during threat modeling. Devin still uses your threat model to rank the findings after the scan.
  • No review pause. The scan cannot stop partway to ask you questions.
  • Threat focus is limited. You can only focus the scan on threats published by Cognition, not on threats your team created. See Focus a scan on threats.
If you need any of these, run a Normal scan.

Threat modeling

A threat model describes what your application does, what it must protect, and how an attacker could reach it. Devin builds one when it scans a repository and uses it to decide which findings matter most. After you select Continue with Devin, Devin opens a session and does the following:
  1. Reads your repositories. Devin maps the entry points, the data the application handles, and the trust boundaries.
  2. Asks you questions. Devin asks about what it cannot learn from the code. For example, which services face the internet, or which data is most sensitive. Answer in the session.
  3. Proposes the scan. Devin shows the threat model and the scan setup. You can ask for changes, such as a focus on certain directories or risks.
  4. Starts the scan. The scan starts after you approve.
If you select more than one repository outside a bulk scan, Devin also decides whether to scan them together or as separate scans. It asks you when it cannot tell from the code. When the scan finishes, Devin ranks the findings against the threat model. A finding that affects something the threat model marks as important ranks higher than the same kind of finding in low-risk code. To check your repositories for one specific risk, such as a newly disclosed vulnerability, use Threats.

Review findings

  1. Open Findings in Security at /security/findings.
  2. Select Open, Reviewed, or Dismissed.
  3. Filter by repository, scan, severity, or pull request state to narrow the list.
  4. Open a finding to inspect its details, affected code, and remediation recommendation when available.
Use the finding’s actions to mark it Reviewed, dismiss it, or return it to Open. You can also select multiple findings for bulk actions. Reviewed is a finding status. It does not mean that a pull request has merged. Use the pull request state to check the fix.

Assign fixes to Devin

Select Fix with Devin in a finding’s details, or select multiple findings and select Fix with Devin. Devin starts remediation sessions for eligible findings. If a finding already has a session, use View session to open it. You can also inspect scan results from the scan session’s Findings tab.

Connect other security tools

Devin Security can import findings from tools such as Snyk, Wiz, Jira, and Linear, and write your triage back to them. The connections you can add depend on your organization’s access. Security connections lists what each connection imports and writes back. For Wiz, see the service account scopes and limits.

Use the API

The Code Scans API guide covers scan creation, status, and findings at organization and enterprise scopes.