How Vizzybl protects your data

Every prompt you send to the Vizzybl AI assistant, and every response it returns, passes through Vizzybl's real-time safety screening layer before it reaches you or the AI model. Screening covers the in-app chat assistant and the Slack assistant; the monitored prompt responses collected by the scrape pipeline are stored as the AI engines returned them and are not passed through this layer.

What we screen for

Vizzybl automatically inspects all interactions across four threat categories:

• Prompt injection and jailbreak attempts — Detects attempts to manipulate AI models into ignoring safety guidelines or producing unauthorized outputs.
• Harmful content — Filters responses containing dangerous, hateful, harassing, or sexually explicit material before they reach your dashboard.
• Malicious URLs — Identifies dangerous links in AI responses so compromised URLs never surface in your results.
• Data leakage — Prevents sensitive information — including personally identifiable information (PII) and proprietary data — from being exposed in AI interactions.

How screening works

Vizzybl's safety layer sits between your prompts and the AI engines. It screens traffic in both directions:

1. Before the AI engine sees your prompt — Your input is inspected for injection attacks and harmful content.
2. Before you see the response — The AI engine's output is screened for harmful material, malicious links, and data leakage.

This two-way inspection runs in real time and does not store the content of your prompts or responses.

Content safety categories

AI responses are evaluated against these safety categories, each with configurable confidence thresholds:

Category	What it catches
Dangerous content	Instructions or promotion of harmful activities
Harassment	Threatening, intimidating, or abusive language targeting individuals
Hate speech	Content promoting discrimination based on protected characteristics
Sexually explicit	Inappropriate sexual content in AI responses

Note: Child safety protections are always active and cannot be disabled.

What happens when a threat is detected

When the safety layer identifies a risk, it takes action based on the severity:

• Low-confidence matches are flagged for review without blocking the response.
• High-confidence matches are blocked before reaching your dashboard.

This approach balances safety with usability — you see clean, relevant AI responses without unnecessary interruptions.

Your data stays private

Vizzybl's screening layer processes interactions in real time without persisting prompt or response content. Your data passes through the safety checks and is not retained by the screening infrastructure.

Next steps

• Set up prompt monitoring — Configure the prompts Vizzybl monitors for your brand
• Understanding prompt results — Learn how to read AI response data in your dashboard