Detect AI Agents Inside Your Web App After Login
A valid account can still drive automated scraping and feature abuse. Use WebDecoy request evidence, actor journeys, and route controls after login.
A real browser opens your app, signs in with a valid account, and starts working through the product. The requests may be authorized. The person who created the account may have handed the browser to an AI agent.
For the application owner, the important questions come next. Is the client walking through every search result? Repeatedly requesting exports? Calling an expensive feature far more often than expected? Which requests should continue, and which need stronger verification or a limit?
WebDecoy helps investigate automation inside a web app through browser signals, server or edge observations, decoy interactions, and actor journeys. Supported application and edge integrations let you apply a response on selected routes. Keep that evidence alongside your application’s authentication and usage records.
A login is the beginning of the investigation
Authentication and authorization still matter. They establish which account made an action and what it may access. Bot detection adds another question: what evidence suggests that the activity is automated?
| In-app activity | Evidence to review | Application context to add |
|---|---|---|
| Repeated searches and pagination | Recorded routes, timing, breadth, and related actor activity | Search action, account reference, and your rate limits |
| Bulk export attempts | Request method, target route, response outcome where recorded | Permission check, export job status, and actual records returned |
| Report or AI-feature requests | Observed automation signals and route decisions | Completed jobs, token usage, account quotas, and measured cost |
| Requests to an isolated decoy | The interaction, its detection, and available identity evidence | Confirmation that the trap exposes no customer data |
An export URL does not prove that an export completed. A denied request does not prove how much money was saved. Match request evidence to the relevant application outcome before making either claim.
Put observation where the activity happens
Add the browser and server components that match your stack. Browser detection can contribute interaction and environment signals when the client executes JavaScript. Server or edge observation covers the request paths you instrument, including clients that never run the browser tag.
For an authenticated app, install coverage on the workspace and sensitive action routes, rather than stopping at the public landing page or login screen. Keep public documentation accessible if you want search engines to index it. Preserve webhooks and service integrations through narrowly scoped access rules and their own verification.
WebDecoy complements Cloudflare and the application’s existing controls. It supplies additional observations and a response workflow where the installed integration supports it.
Read the journey behind the alert
Open Actors, select an actor, and open Journey. The observed journey connects stored requests to routes, origins, and available signals. Its chronological list lets you inspect the individual detections. Saved AI journey insights can summarize likely collection targets and access patterns with links back to supporting requests.
Our published journey examples show the distinction: one sample repeatedly requested a single article, while another covered many different targets. They demonstrate request-pattern investigation on our public site. They are not evidence of abuse inside a customer’s authenticated app.
The identity boundary matters. An actor based on network evidence may group more than one client. An inferred session marks a gap in observed requests; it is not your app’s login session. Use trusted application records to correlate a pseudonymous account reference, action, time, and request identifier where available. Avoid sending passwords, access tokens, raw prompts, or customer records as investigation context.
Choose the response for the action
Start with a route in monitor mode and inspect its decisions. Then choose supported controls for the action: stronger verification, a scoped access policy, or SDK rules alongside application quotas and authorization.
Clearance enforcement distinguishes a browser check, positive human signals, and a verified challenge. A valid browser clearance token alone does not establish that a human is driving the browser. Confirm which verification requirements your deployed validator applies before relying on them.
Decoys can connect observation to a later response. When the required client and edge integration is installed, a same-origin decoy or supported SDK tripwire can associate the interaction with clearance revocation. A shared-domain detection-only decoy cannot receive your site’s first-party clearance cookie and cannot perform that revocation. Inspect setup status rather than assuming every trap has the same effect.
Verify the whole loop on a test workspace
Use an isolated workspace with synthetic data and a bounded request budget. A useful after-login demonstration should establish:
- The ordinary browser workflow succeeds with the intended permissions.
- An authorized AI-driven browser signs in and carries out a small, predefined task.
- WebDecoy records the observed activity and any deliberately included decoy interaction. State whether the agent discovered the trap or was explicitly directed to it.
- The application audit log connects those requests to the synthetic account. Preserve uncertainty where a join is unavailable.
- Monitor-mode decisions remain separate from actual refusals. If enforcement is enabled, confirm that the protected handler did not run when a request was refused.
- An ordinary follow-up workflow still succeeds, and required integrations remain available.
That is the acceptance test for an authenticated demonstration, not a claim that this article has completed one. Report missing coverage and failed checks beside successful ones. One controlled run establishes its own result, rather than a detection rate for every AI agent.
Start with the feature you need to understand
Choose a search, export, report, or API workflow. Install the appropriate sensor, inspect the evidence, and correlate it with your application’s activity. Then configure a response for that route and verify the actual outcome.
Install WebDecoy in your app and review the web-app protection workflow.
Frequently Asked Questions
Can a bot use a valid account in a web app? +
Yes. Authentication establishes the account and its permissions; it does not establish that a person manually drives every later request. Observe activity and enforce appropriate controls on sensitive actions after authentication.
Does WebDecoy automatically identify my logged-in users? +
No. A WebDecoy actor groups traffic using available identity evidence. Correlate observed requests with your application's trusted audit logs to establish customer and account context. Inferred request sessions are not authenticated user sessions.
Does installing a browser script block AI agents? +
A browser script provides browser-side evidence. Blocking requires a configured application or edge integration and an enforced policy. Monitor mode allows requests, and default browser clearance does not prove human identity.
Can WebDecoy work alongside Cloudflare? +
Yes. WebDecoy adds browser and application observations, decoy evidence, and supported route controls alongside Cloudflare. Keep existing authentication, authorization, rate limits, and appropriate edge controls.
Continue your investigation
- Install in your web app
Choose the browser and server components for your framework.
- Review the web-app workflow
Connect app activity, automation evidence, and supported responses.
- Inspect an actor journey
See how recorded requests become an evidence-backed investigation.
Share this post
Like this post? Share it with your friends!
Want to see WebDecoy in action?
Get a personalized demo from our team.