Glossary

The following terms appear in the extension, the dashboard, and this help center.

People and access

Access level: What someone can do in Sanitized AI: Member, Dashboard only, or Admin. See Roles and access levels.
Member: Someone the browser extension protects. Members don’t have a dashboard unless people report to them.
Dashboard only: Someone who reviews the activity of the people who report to them, without using the extension. Doesn’t use a seat.
Admin: Someone who manages the organization’s people, rules, and settings. Admins use the extension too.
Reports to: The setting that says who someone’s manager is. Anyone with people reporting to them gets a team dashboard.
Activation: The step where an admin makes an account usable. It sends the welcome email. Before that, the person is Not activated and can’t sign in.
Seat: A paid place in your plan. Each active Member and Admin uses one.

Sign-in link: A single-use link that Sanitized AI emails you. It expires after 15 minutes.
Single sign-on (SSO): Signing in with your company account, such as Google, Microsoft, or Okta.
SCIM: A standard that lets your identity provider create and disable Sanitized AI accounts automatically.
Managed extension: A version of the extension that your IT team installs on the computers it manages.
Organization ID: Your organization’s identifier. IT uses it to deploy the managed extension. Nobody types it to sign in.

Guardrail: A rule that recognizes one kind of sensitive data, such as credit card numbers. See Guardrails.
Naming rule: A guardrail an admin builds from the way the organization names its files, such as its matter numbers. It flags files by their name and finds the reference number in prompts. See Naming rules.
Severity: How much harm the data a guardrail catches could cause, from Low to Critical.
Nudge: An enforcement action that shows a Security Check warning, with the option to send anyway.
Block: An enforcement action that shows a Blocked by Policy warning. The prompt can’t be sent, or the file attached, as is.
Monitor: An enforcement action that records the event without showing a warning.
Soft launch: A block that starts as a nudge for 14 days, and then becomes a block on its own. See Soft launch a block.
Exception: An exemption for a person or department from some guardrails, with an expiry and a reason. See Exceptions.
Warning Banner Note: A short explanation an admin writes for a guardrail. It appears in the warning.
Sanitize: Replace the sensitive parts of a prompt with safe stand-ins before sending it.
Fake data: Realistic, made-up stand-ins, such as an invented name, used instead of placeholders when you sanitize a prompt.
False positive: A flag on something that isn’t actually sensitive. You can report it from the warning.
Allowlist: Words and phrases your admins have marked as never to be flagged.

Guarded site: An AI tool that Sanitized AI protects for your organization. See Guarded sites.
Report card: Sanitized AI’s risk assessment of an AI tool, with a score out of 10.
Risk-aware enforcement: A setting that turns warnings into blocks on riskier AI tools.
Event: A record that a prompt or file was flagged. It doesn’t contain the prompt’s text. See Events.
Outcome: What someone did after a flag: sanitized, re-edited, overridden, ignored, or reported as a false positive.
Override: Sending a flagged prompt with Send Anyway.
Risk score: A 0 to 100 summary of how severe your organization’s or team’s flagged events are.
Compliance profile: Where your organization operates and what it does. It decides which laws, codes and standards the Audit Assistant and the AI Safeguards Report cover.
Audit Assistant: The card on the Compliance tab that shows where your AI safeguards stand against each law and framework in your profile, and the gaps to fix first. See Compliance and the Audit Assistant.
AI Safeguards Report: A PDF that maps what Sanitized AI did, and your numbers for a period, to the laws and frameworks in your profile. It’s evidence for your own obligations, not a certificate of compliance.