Solutions
Shadow AI apps, and a sanctioned path for them
Shadow AI apps are business apps that employees build with AI coding tools and publish outside IT's sign-in, review, hosting and audit. They often run on personal accounts with public addresses, copied credentials and no release review. Diakrin gives these apps a sanctioned path instead of a ban: employees keep their AI tools, while apps run in the company's own cloud behind company sign-in, with IT-approved data access, approved releases and budgets.
Last reviewed
Shadow AI apps are business apps that employees build with AI coding tools and put into use outside IT's sign-in, review, hosting and audit. They exist because the tools work: an employee can now build a working request tracker or dashboard in an afternoon. The problem is not the app. It is where it runs, what it can reach and that nobody approved it. Diakrin gives these apps a sanctioned path instead of a ban.
What are shadow AI apps?
Shadow IT used to mean unapproved SaaS subscriptions. With AI coding agents and AI app builders, employees now create their own software. A typical shadow AI app is built by a business team, deployed on a personal or free hosting account, shared with colleagues by link and connected to company data with whatever credentials the builder had at hand.
Security teams usually find these apps late, because they live outside the company's identity provider, cloud accounts and source control.
What are the security risks of vibe-coded apps?
The risks are mostly about the path to production, not the code style:
- No company sign-in. Apps on public addresses can be opened by anyone who finds the link.
- Credentials in the wrong place. Keys and passwords pasted into prompts, source files or environment settings on personal accounts.
- Company data on third-party hosting. Copies of business records in databases the company does not own or see.
- No release review. Nobody checked what changed before colleagues started relying on it.
- Agents with too much reach. An agent with broad credentials can act beyond what the task needs. OWASP lists excessive agency among its top risks for LLM applications ↗.
- No owner, no inventory. When the builder leaves, nobody knows what data the app holds or who uses it.
Why does banning AI coding tools not work?
The demand comes from real work that IT cannot build fast enough. A ban moves that work out of sight, which makes it harder to govern. Frameworks such as the NIST AI Risk Management Framework ↗ treat AI risk as something to govern, map, measure and manage. For internal apps, that means a supported path that is easier than the unsanctioned one.
How do you bring shadow AI apps under IT control?
Give builders a path that keeps the speed and adds the controls. In Diakrin, each risk above has a specific answer:
| Risk | What Diakrin does |
|---|---|
| Public, unauthenticated apps | Employees open apps with company sign-in, and each app has its own audience. On Google Cloud, apps stay on your private network behind Identity-Aware Proxy. |
| Credentials in prompts | Credentials are entered by an authorized owner on a customer-hosted form, never in the coding tool. |
| Data on third-party hosting | Each app's database, files and secrets stay in your cloud account. |
| No release review | Every production release needs an approved plan, from an eligible reviewer who is not the requester where your policy requires it. |
| Agents with too much reach | The agent acts as the person who connected it, and every call is checked on the server. |
| No inventory | IT sees every app's cost, data and activity in one place. |
See Manage access and security.
How it works with Diakrin
- Offer the sanctioned path. IT connects company SSO, creates a company cloud setup in AWS or Google Cloud and decides who may build. See Get started.
- Bring existing apps in. Builders can start fresh or bring an app from their coding tool or a company repository. Imported code is assessed without running it. See Preview, approve and deploy.
- Replace copied credentials with grants. The app requests named data operations. IT approves each with resource and field limits, and the app and each employee get separate grants.
- Review before anything goes live. The release plan names the exact image, target, cost estimate and access implications, and a reviewer approves or denies it.
- Keep watching. IT tracks health, cost, data and activity, and can change an app's audience, quarantine its data or stop its runtime. See Review capabilities and deployment.
What should IT do first?
Start with one team that already builds with AI and one small app with a clear owner, such as a request tracker. Use synthetic data until real data access is approved. Then read AI app governance and IT approval for the review checklist IT can apply to every app after that.
Frequently asked questions
What is a shadow AI app?
A shadow AI app is an app an employee builds with an AI coding tool and puts into use without the company's sign-in, review, hosting or audit. It is the AI-era form of shadow IT.
What are the main security risks of vibe-coded apps?
The common ones are public addresses with no company sign-in, credentials pasted into prompts or source, company data copied to personal accounts or third-party hosting, no review of what was released, no owner when the builder leaves, and no record of what data the app holds.
Should IT block AI coding tools?
Blocking tends to move the work out of sight. A sanctioned path lets employees keep the tools they use while IT controls where apps run, what data they reach, who can open them and what goes live.
How does Diakrin keep credentials out of the coding tool?
Company credentials never pass through a coding tool. An authorized owner enters them on a customer-hosted form, and the app and the employee each still need an approved grant before the app can use them.
Can IT see the data an AI-built app holds?
On AWS, the Diakrin runner in your cloud reports metadata such as database and table sizes, estimated row counts and column names, file counts and secret rotation age. Row values, file contents and secret values are never collected. This report is not yet available on Google Cloud or self-hosted installations.
What happens to apps when their builder leaves?
Removing the person from the Diakrin organization revokes their app and project grants, ends their recorded app sign-in sessions and stops their coding-tool tokens. IT can reassign the app's technical owner and data owner.
Sources
Facts about other products come from their public pages, checked on the dates shown. Products change; tell us if something is out of date.
- genai.owasp.org/llm-top-10 (checked 9 November 2026)
- nist.gov/itl/ai-risk-management-framework (checked 9 November 2026)
See Diakrin with your own cloud
Let your team build internal apps with the AI tools they already use, while IT keeps control of identity, data, approvals and spend.