Filter Policies

Filter policies define content filtering rules that are applied to AI input and output. They can block or redact specific words, patterns, and sensitive data in AI conversations. Navigate to AI Services > Filter Policies in the dashboard.

Creating, editing, and toggling policies requires the manage AI governance permission — users without it see policies in read-only mode.

Overview

Filter policies contain one or more rules, each of which matches content using word lists or regular expressions and takes an action (block or redact). Policies can be:

• Builtin — pre-configured policies provided by Quant (read-only)
• Custom — policies you create and manage

Policies can be assigned to individual agents, or enforced organisation-wide through AI Governance.

Managing policies

The filter policies list shows all policies with their type, rule count, action type, and enabled status. Use the search bar and type filter to find specific policies.

Column	Description
Policy	Name and description
Type	Builtin or Custom
Rules	Number of filtering rules
Action	BLOCK, REDACT, or MIXED (if rules use different actions)
Enabled	Toggle to enable or disable the policy

Click the toggle to enable or disable a policy. Disabled policies are not evaluated during AI requests.

Creating a policy

Click Create Policy to open the policy editor.

Policy details

Field	Description
Name	A descriptive name for the policy (required)
Description	What the policy does

Rules

Each rule defines a pattern to match and an action to take. Click Add Rule to create a new rule.

Field	Description
Rule Name	A label for the rule (e.g. “Block profanity”)
Match Type	Word for exact keyword matching, or Regex for pattern matching
Words / Pattern	Comma-separated word list (word mode) or a regular expression (regex mode)
Action	BLOCK rejects the entire request; REDACT masks the matched content
Apply To	Both (input and output), Input only, or Output only

Word matching is case-insensitive and uses word boundaries, so matching “test” won’t match “testing”. Regex matching supports full regular expression syntax.

Rules are evaluated in order — the first BLOCK match rejects the request immediately.

Testing a policy

The policy editor includes a built-in Test Policy panel that runs entirely in your browser (nothing is sent to the server). To test:

1. Expand the Test Policy section
2. Paste sample text into the text area
3. Select whether to test as Input or Output
4. Click Test Rules

The test shows:

• Verdict — PASSED or BLOCKED with the reason
• Highlighted text — matched content highlighted with BLOCK (amber) or REDACT (red) markers
• Match summary — table of all matches with rule name, action, matched text, and match type

Example use cases

Block profanity

• Match Type: Word
• Words: list of prohibited terms
• Action: BLOCK
• Apply To: Both

Redact Australian phone numbers

• Match Type: Regex
• Pattern: \b\d{4}\s?\d{3}\s?\d{3}\b
• Action: REDACT
• Apply To: Output only

Block competitor mentions

• Match Type: Word
• Words: competitor names
• Action: BLOCK
• Apply To: Input only

Applying policies

Filter policies can be applied at two levels:

1. Per-agent — assign policies in the agent editor’s Delegation tab or the Slack bot editor. Agent-level policies are combined with any org-wide mandatory policies at runtime.
2. Organisation-wide — enforce mandatory policies through AI Governance so they apply to every AI request regardless of agent settings

Next steps

• AI Governance — Enforce policies organisation-wide
• Agents — Assign filter policies to individual agents
• Slack Bots — Apply filter policies to Slack bot agents