# AI Reading Detection

- **Spot likely off-screen assistance**  
- **Detect reading from a second monitor, tablet or phone with AI tools feeding answers**  
- **Time Stamped evidence**

## Why this matters

Generative AI and online resources are now part of many candidates’ workflows. That’s not always a problem—but when the goal is to evaluate _their_ skills under realistic constraints, silent assistance can skew results. Reading Behavior Analysis helps teams:

- **Spot likely off-screen assistance** (e.g., a second monitor, tablet, or phone) even when no on-device plugins are involved.
- **Triage submissions faster** with structured, time-stamped evidence.
- **Protect fairness** for candidates who play by the rules.

## What reviewers see

- **Short summaries** (e.g., “Reading pattern detected for ~18s before answer to Q2”).
- **Contextual overlays** with the relevant metrics (eye path stability, head pose variance, relative speed signals) so reviewers can quickly judge severity and intent.
- **Consolidated report** across the full session to understand whether signals are isolated blips or a recurring pattern.

## AI usage detection—what we claim (and don’t)

Reading Behavior Analysis is _not_ a polygraph or a definitive AI detector. Instead, it flags **behavior consistent with reading** from an external source while answering. That can include:

- A second monitor with AI tools or web content
- A phone or tablet just off camera
- Printed notes or another screen placed beside the camera

If it matches a reading pattern, Feenyx proactively surfaces it for review— **regardless of device type**.

## FAQs

**Does this affect candidate experience?**

No changes to the recording flow. Analysis runs post-upload.

**What about false positives?**

Reviewers remain the arbiter. You can mark and exclude segments, adjust sensitivity, and whitelist allowed behaviors (e.g., reading problem statements on-screen).

**Can it catch AI usage for sure?**

It flags _reading-like behavior_ strongly associated with using external content—including AI tools—but it’s not a definitive AI detector. Treat it as high-signal evidence, not a verdict.

**What data is stored?**

We retain the original video (per your retention policy) and derived motion metrics tied to the session for audit and model improvement within Feenyx’s privacy framework.

**Is this inclusive for candidates with different needs?**

Yes. We support policy exceptions and reviewer notes. If your process allows accommodations (e.g., reading aids), you can reflect that in policy and reporting.