Our Approach to Content Safety

Our principles

Content safety in generative AI is not a solved problem. Every platform makes tradeoffs between creative expression and harm prevention. Here's how we think about it:

1. Default to allowing creative expression. Most AI-generated content is harmless. Our system should not over-block legitimate creative use cases.
2. Prevent real-world harm. Content that depicts or could facilitate violence against real people, CSAM, or non-consensual intimate imagery is never acceptable.
3. Be transparent about decisions. When we block content, we tell users why. When we change policies, we explain the reasoning.
4. Iterate based on evidence. Our moderation system improves continuously based on real data, not hypothetical concerns.

Multi-layered moderation

We use a three-layer approach:

• Input screening — Text prompts and uploaded images are checked before generation begins. Known harmful patterns are blocked immediately.
• Generation-time guardrails — The AI models themselves have built-in safety training that prevents generating certain categories of content.
• Output review — Generated content passes through a classifier before delivery. Content flagged as potentially harmful is held for review.

What we block

Our content policy prohibits:

• Content depicting minors in any inappropriate context
• Non-consensual intimate imagery of real people
• Content that promotes violence or terrorism
• Deepfakes intended to deceive or defraud

We do not block artistic nudity, fantasy scenarios, or creative content that doesn't target real individuals. We believe adults should have broad creative freedom with AI tools.

Reporting and enforcement

Users can report content via the dashboard or by emailing safety@canvaslabs.us. We review all reports within 24 hours. Confirmed violations result in content removal and, for serious or repeated violations, account suspension.

Transparency report

Starting Q2 2026, we will publish quarterly transparency reports covering moderation volume, categories of blocked content, appeal outcomes, and policy changes. We believe accountability requires data, not just promises.