Content Moderation
Every post on Circus passes through layered checks before it reaches anyone's screen — hash matching, machine learning, and human review. Here's how it works.
Three layers
Every post passes through three lines of protection before it reaches anyone's screen.
Known harmful content never gets in
Before anything is stored, every upload is checked against databases of known illegal imagery. Exact-match text violations are caught instantly.
AI reads every post before it goes live
Models scan text, images, and video in parallel. High-confidence violations are blocked automatically; borderline cases go to a person.
Real people make every tough call
Our Trust & Safety team handles anything the automated systems can't confidently decide. No permanent action is taken by a machine alone.
Automated detection
These systems run on uploads and published content — together, not instead of each other.
AWS Rekognition
- Detects nudity, graphic content, violence, and visually explicit material
- Runs on images and video thumbnails before content goes live
- High-confidence hits are blocked; borderline results queue for humans
AWS Comprehend
- Flags hate speech, toxic language, and PII in posts, comments, and bios
- Works across multiple languages
- Feeds risk scores into the review queue alongside member reports
Circus-trained ML
- Trained on platform-specific patterns — satire vs. harassment, creator context
- Retrained on cases our reviewers have decided
- Informs prioritization; never issues final enforcement alone
Known-violation database
- Compares image and video uploads against law-enforcement CSAM hash lists
- Runs before publication — matches never reach the feed
- Triggers immediate hold, account suspension, and mandatory reporting
Human review
Automated systems flag likely violations at scale. People make the final call.
- Reviews complex, borderline, and high-stakes cases
- Confirms every permanent ban before it takes effect
- Trained on our Content Guidelines and community policies
- Operates around the clock with priority queuing for urgent safety content
Response time targets:
(risk to life)
Content during review
Flagged or reported content enters one of three states while we decide.
Visible, under review
Most flagged content stays visible while we review. Most reports do not result in removal.
Visibility-restricted
When context matters, we may reduce discovery reach without removing the post while we review.
Held pending review
CSAM, graphic violence, or known terrorist imagery is held before publication and reviewed first.
Creators are notified when their content enters review and when a decision is made.
Child safety
Zero tolerance for CSAM
Hash-matching runs on every image and video upload, comparing against law enforcement databases of known child sexual abuse material.
- Immediate content block — the material is never published
- Automatic account suspension pending review
- Mandatory report to relevant authorities, including NCMEC's CyberTipline as required by law before user uploads go live
Content that appears to sexualize minors — even without a hash match — is escalated to specialist reviewers and to the relevant authorities.
Creator tools
Creators manage their communities before anything reaches the Circus safety team.
Pin & remove
Pin posts to keep important content visible. Remove any post from your community instantly — no queue.
Warn, restrict, ban
Issue formal warnings, restrict posting privileges, or remove someone from your community permanently.
Keyword filters
Block comments containing specific words or phrases. Set controls per post or members-only commenting.
Transparency
We plan to publish transparency reporting on content removed, reports received, government requests, and enforcement actions. Published statistics will live on the safety hub when available — we do not commit to a fixed publication cadence before launch.
Urgent safety issues
For content involving imminent risk to life, contact us directly. Do not wait for the in-app report flow in urgent situations.