Trust Center
The answers your security review needs
What Sanitized Ai commits to, how it maps to the obligations you carry, and how to reach us. One page you can send to your security team.
Our commitments
Prompt privacy by design
Administrators see flagged event metadata only: category, count, tool, time. Prompt content is not stored or shown, unless an employee chooses to report a false alarm. Then only the flagged text and up to about 400 characters around it are kept for review.
Control before submission
Detection and redaction act before a prompt reaches the AI provider, the only point where control is still possible.
Audit-ready by default
Every policy event is logged, and reports export for auditors, clients, and regulators. Every guardrail change is recorded too, with a tamper-evident copy, and admins get an email whenever protection is weakened.
Deploys the way IT already works
Push the extension to Chrome, Edge, Firefox, and Brave through browser policy in Google Workspace, Intune, or similar tools, with no network changes. Start in monitor mode, then turn on warnings and blocks one guardrail at a time.
99.9% uptime target
The service is built to stay out of your team's way. This is an operating target we monitor, not a contractual SLA.
Canadian company
Sanitized AI Inc. is incorporated in Canada and built fluent in Canadian privacy law. Detection works on French prompts as well as English, and employees can see warnings in French.
No third-party AI sees your prompts
Detection runs on models we host ourselves, never on a third-party AI service.
Files never touch disk
Uploaded files are processed in memory and never written to disk, in a detection service isolated for each customer.
Fits your identity stack
Single sign-on with Microsoft, Google, and Okta, SCIM user provisioning, and separate roles for admins, managers, and users.
Mapped to the obligations you carry
Sanitized Ai is a control you deploy to help meet your own obligations. Here's where it fits:
Data-handling questions, answered
Can Sanitized Ai read our employees' conversations?
No. The admin view shows flagged event metadata only: category, count, tool, and time. Prompt content is not stored or shown to anyone. The one exception is a false alarm an employee chooses to report: then only the flagged text and up to about 400 characters around it are kept, so the detection can be reviewed.
When does redaction happen?
Before the prompt reaches the AI provider. Once data is submitted to a public AI tool it can be retained, sub-processed, or used for training, so the control acts at the last moment it still can.
Will you complete our security questionnaire?
Yes. Send it to sales@sanitized.ai and we'll return it along with our privacy documentation. We typically respond within 24 hours.
Which AI tools are covered?
ChatGPT, Claude, Gemini, Copilot, and Perplexity, along with the other AI tools shadow-AI detection surfaces. Coverage grows as new tools appear, so we keep a current list rather than a fixed number.
Which file types are covered?
File uploads are scanned in the same pass as prompts: DOCX, XLSX, PPTX, and PDF. In Word and PowerPoint files, speaker notes and embedded images are checked too.
Responsible disclosure
Found a vulnerability? We want to hear about it. Email sales@sanitized.ai with the details and we'll route it to the right people. First response is typically within 24 hours.