# Cheat Detection

> HeyMilo's Cheat Detection helps recruiters identify potential cheating behaviors during candidate interviews. This guide explains how to configure the system and interpret the results.

## System Requirements

Before enabling cheat detection, ensure:

* Cheat detection is enabled for your workspace
* Candidates use a supported browser for web interviews
* Camera and microphone permissions are granted

If cheat detection is not available, contact [**support@heymilo.ai**](mailto:support@heymilo.ai).

<details>

<summary><strong>Supported Browsers</strong></summary>

**Recommended (best experience)**\
We recommend using one of these browsers, kept up to date:

* Google Chrome: Latest stable version. Best tested; recommended for the most reliable experience.
* Microsoft Edge: Latest stable version. Full support and regularly tested.
* Mozilla Firefox: Latest stable version. Full support.
* Safari: Latest version on macOS and iOS. Use the latest Safari your device supports.

**Also supported**\
These browsers use the same or similar engines as the recommended ones and should work when up to date:

* Opera: Current stable version (Chromium-based).
* Opera GX: Current stable version (Chromium-based).
* Brave: Current stable version (Chromium-based).
* Samsung Internet: Recent versions on Android (Chromium-based; keep the app updated).
* Vivaldi: Current stable version (Chromium-based).
* Arc: Current stable version (Chromium-based on macOS/Windows).

**Not supported**

* Internet Explorer (all versions): Not supported; use Edge or another modern browser.
* Very old or unpatched browsers: Browsers that are no longer receiving security updates are not supported.

**General guidance**

* Desktop: Chrome, Edge, or Firefox on Windows/macOS/Linux, or Safari on Mac.
* Mobile: Safari on iOS or Chrome/Firefox/Samsung Internet on Android.
* Keep your browser updated to the latest version for the best experience.

</details>

## Where Cheat Detection Is Configured

Cheat detection can be configured in **two places**, depending on your workflow.

### Option 1: Workspace-Level Settings (Recommended Default)

Use this if you want **consistent integrity settings across all interviews**.

<figure><img src="/files/B7wYm8evNEQVtcGrkiji" alt=""><figcaption></figcaption></figure>

**Path:**

**AI → Interview Integrity**

#### Settings Available

* **Enable Cheat Detection**

  Turns integrity monitoring on for all new interviews in the workspace.
* **Detection Threshold (0–100)**

  Controls how strict the system is when flagging events.

  * Lower values = less sensitive (fewer flags)
  * Higher values = more sensitive (flags more signals)
  * Recommended starting point: **70**
* **Detection Types**

  Select which signals to monitor:

  * Facial Behaviour
  * Multiple People
  * Phone Detection
  * AI / Scripted Answer
  * Unusual Delays
  * Tab Switching

Click **Save Settings** to apply changes.

### Option 2: Interview-Level Settings (Per Agent)

Use this if you need **role-specific control**.

<figure><img src="/files/3lv1tD9E4ypYWqVaBH2E" alt=""><figcaption></figcaption></figure>

**Path:**

**Create or Edit Interviewer → Voice/Video Interview (workflow step) → Settings tab**

You’ll see the same options as workspace-level settings:

* Enable Cheat Detection
* Detection Threshold
* Detection Types

These settings apply **only to that interviewer** and override workspace defaults.

## What Each Detection Type Means

These signals are **observational**, not judgments.

<figure><img src="/files/FO5GwKY4hCKWV6NR3WzE" alt=""><figcaption></figcaption></figure>

<details>

<summary>Facial Behaviour</summary>

Flags repeated or sustained face movement away from the screen that may indicate reading notes or external prompts.

</details>

<details>

<summary>Multiple People</summary>

Detects the presence of more than one person in frame during the interview.

</details>

<details>

<summary>Phone Detection</summary>

Flags when a mobile device appears in view during the interview.

</details>

<details>

<summary>AI / Scripted Answer</summary>

Identifies response patterns that strongly resemble pre-written or AI-generated text, based on structure and delivery timing.

</details>

<details>

<summary>Unusual Delays</summary>

Flags long or inconsistent response delays that may suggest off-screen assistance.

</details>

<details>

<summary>Tab Switching</summary>

Detects when the candidate navigates away from the interview window during a response.

</details>

## Viewing Cheat Detection Results

**Path:**

<figure><img src="/files/UXShwrWkMIotzPpFyZXi" alt=""><figcaption></figcaption></figure>

**Interviewers → Candidate → Diagnostics & Analysis**

Scroll down to Cheat Detection Overview and you’ll see:

* Events grouped by detection type
* Timestamps for each flagged moment
* Confidence scores per event
* A synced player to jump directly to flagged points

<details>

<summary>Understanding Confidence Scores</summary>

Confidence scores indicate **how strongly a signal was detected**, not intent.

* **80–100**: Strong signal, review recommended
* **60–79**: Moderate signal, check context
* **Below 60**: Likely environmental or accidental

**Always review video and transcript context before drawing conclusions.**

</details>

<details>

<summary>Patterns Matter More Than Single Events</summary>

More weight should be given to:

* Repeated signals across the interview
* Multiple detection types occurring together
* Signals aligned with complex or high-stakes questions

Single, low-confidence events are often caused by lighting, camera angle, or natural movement.

</details>

<details>

<summary>What Cheat Detection Does <em>Not</em> Do</summary>

* It does **not** automatically fail or reject candidates
* It does **not** score candidates higher or lower
* It does **not** analyze facial features, expressions, accent, or appearance
* It does **not** make hiring decisions

These signals exist to **support review**, not replace judgment.

</details>

<details>

<summary>Best Practices</summary>

* Start with a threshold around **70**
* Enable only the detection types you actually plan to review
* Always pair integrity signals with interview content
* Document how your team interprets signals
* Revisit settings as roles or hiring volume change

</details>

<details>

<summary>Troubleshooting</summary>

**No cheat detection data**

* Feature may be disabled
* Interview may have been too short
* Camera permissions may not have been granted

**Too many flags**

* Decrease detection threshold
* Review candidate setup instructions

**Inconsistent results**

* Check browser compatibility
* Review lighting and camera placement

</details>

{% hint style="success" %}
Example: A candidate with consistent 100% confidence “Eye Movement” detections from 1:20 to 8:03 likely looked away repeatedly, possibly at a phone or notes.
{% endhint %}

## Red Flags vs. False Positives

<details>

<summary>💡 <strong>High Risk</strong></summary>

Multiple detection types, repeated high-confidence incidents, aligned with tough questions.

</details>

<details>

<summary>💡 <strong>Possible False Positives</strong></summary>

Single low-score events, poor lighting, setup/connection issues.

</details>

## Additional resources

If you need help enabling or configuring cheat detection, contact <support@heymilo.ai> or use in-app chat.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.admin.heymilo.ai/creating-your-interview-agents/setup/cheat-detection.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.