See what your
moderation is missing
A Tox Scan analyzes a sample of your moderated data to reveal harmful behavior patterns that may be slipping through everyday moderation.
Tox Scan uses the same AI models that power Amanda, Aiba’s moderation platform used by online games and social communities to manage millions of conversations and messages.
The initial scan and findings summary are provided at no cost.
Moderation always
has blind spots
Most platforms that run a Tox Scan believe their moderation is working reasonably well. The scan usually confirms that, mostly. And then it shows them exactly where it is not.
The charts show a typical finding. Input data, what existing moderation already flagged, sits at 22%. After deeper analysis that number rises to 47%. More than twice as much harmful content was present in the same dataset.
That gap reflects how quickly language evolves and how effectively bad actors learn to stay just below the threshold of what filters catch. Misspellings, coded phrases, and subtle escalation rarely look alarming on their own. Across hundreds of conversations, the pattern becomes hard to ignore.
How the Tox Scan works
Upload a sample dataset
Share a sample of moderated chat messages, posts, or reports from your platform using a secure upload link we provide after you request the scan.
A dataset representing a typical day or week of activity is usually enough to surface meaningful patterns. Upload what best reflects how your platform operates.
AI analysis
Aiba’s AI models analyze the conversations for harmful behavior, language patterns, and interaction signals.
Findings summary
You receive a short findings summary highlighting patterns, signals, and harmful interactions that may be slipping through. The initial scan and report are provided at no cost.
What your Tox Scan reveals
Your report highlights patterns that are difficult to see during everyday moderation.
After the analysis you receive a short findings summary based on the dataset you provided.
The report typically highlights:
- Hidden toxic content your system did not detect
- Percentage of violations missed by current moderation
- Most common harmful behavior patterns in your community
- Risk benchmark compared to similar platforms
- Signals that indicate emerging moderation risks
I approved the scan mostly to rule things out.
The findings ended up directly informing our roadmap for the next two quarters.
VP of Operations, Major US Game Studio
Why not just run this through a general AI?
You could take a sample of community messages and run them through a general AI model. It will find some harmful content.
But a general model has no baseline. It does not know what harmful behavior looks like inside a gaming community at 11pm on a Friday.
It cannot recognize the coded language particular groups use to harass without triggering keyword filters.
Aiba’s models are trained specifically on community moderation data.
They understand the difference between competitive trash talk and targeted harassment. They recognize the slow drift toward toxicity that often precedes incidents.
That context turns detection into useful moderation insight.
Secure analysis
using a limited dataset
Secure analysis
using a limited
dataset
Aiba is a moderation infrastructure company. Your data is used for one purpose: to produce a useful report for you. It is never used to train models, shared with third parties, or retained beyond the analysis.
After requesting the scan you receive a secure upload link where you can share a limited sample dataset with us.
NDA agreements are available.
FAQ →
Need a deeper investigation?
The X-RAY report
A deep dive into your social data
For platforms that want a more detailed analysis of community behavior, the X Ray Report offers a deeper review of moderation patterns, risk signals, and platform dynamics.
This analysis examines larger datasets and provides a more comprehensive view of how harmful behavior spreads through conversations.
