The Security & Fallback Settings panel on the Configure tab controls how an AloAi Voice Agent handles unsafe content, what happens if its voice provider fails during a call, and how much call data is kept and for how long.
This article covers safety guardrails, the fallback voice, data storage and retention, and redacting personal information (PII, personally identifiable information) from call data.
To get there, open the Agents menu and click the agent’s name. The Security & Fallback Settings panel is on the right side of the Configure tab.
Open the agent from the Agents list.
The Security & Fallback Settings panel on the Configure tab.
The Security & Fallback Settings panel.
Safety guardrails
Safety guardrails are an optional content-moderation layer. When turned on, they check content in real time on both sides of the conversation: what the caller says and what the agent is about to say. The check runs before the agent responds or triggers an action, so flagged content is caught before it reaches the caller.
Guardrails are off until you set them up. Each category is a separate checkbox, so you can turn on only the ones that fit your use case. Standard, on-topic conversations are not affected.
Turning on guardrails
In the Security & Fallback Settings panel, find the Safety Guardrails card and click Set Up.
Select the categories you want to enforce under Input Guardrails and Output Guardrails, then click Save. You can turn on any combination.
Changes apply to new calls without restarting the agent. Calls already in progress are not affected.
The Safety Guardrails window.
Output categories
Output guardrails check what the agent is about to say before it is converted to speech:
Harassment: harassment or abusive language
Self-harm: self-harm or suicide-related content
Sexual Exploitation: sexual exploitation or explicit content
Violence: violence or threats of harm
Defense & National Security: defense or national security-related topics
Illicit & Harmful Activity: illegal or harmful activities
Gambling: gambling-related content
Regulated Professional Advice: regulated professional advice, such as medical or legal guidance
Child Safety & Exploitation: child safety or exploitation-related content
The Output Guardrails categories.
Input category
The input guardrail checks what the caller says before it reaches the AI model:
Platform Integrity (Jailbreaking) detects attempts to bypass the agent’s instructions, override its behavior, or get around its safety rules through prompt injection or similar manipulation.
A flagged input is blocked before the model sees it and is left out of the call’s conversation history, so it cannot influence later responses.
The Input Guardrails category.
When a guardrail is triggered
If the caller’s message is flagged (input guardrail), the agent does not process it and replies with a short, safe response instead. The flagged message is not kept in the conversation, so the agent does not act on it later in the call.
If the agent’s reply is flagged (output guardrail), the agent does not say it and gives a short, safe response instead.
The call continues. Guardrails do not end or transfer the call. Each trigger is logged with the category, the direction (input or output), and a timestamp. You can review these entries in the call’s history.
Fallback voice
A fallback voice keeps a call going if the agent’s voice provider stops responding. When the agent detects a voice provider failure, such as a connection error, timeout, or provider error, it switches to a fallback voice and the conversation continues.
Under Fallback Voice ID in the Security & Fallback Settings panel, choose a mode:
Automatic fallback is the default. When the primary voice fails, a voice of the same gender from a different voice provider is used. No other setup is needed.
Select a fallback voice lets you choose one or more fallback voices. Each one must come from a different voice provider than the primary voice and than every other fallback voice in the list. You can add, remove, and reorder them.Fallback Voice ID with the two modes.
Fallback Voice ID with the two modes.
The Select Voice window.
How fallback works on a call:
Fallback voices are tried in the order listed. If every voice in your list also fails, automatic fallback is used last.
Once a fallback voice takes over, the agent keeps using it for the rest of the call, even if the original provider recovers.
Each time a fallback voice is used, the event is recorded on the call and shown in call history.
Saving fallback settings does not change the agent’s primary voice.
Selected fallback voices, tried in the order listed.
Data storage and retention
Data Storage and Retention Period control how much of the agent’s call data is kept and for how long. Deletion based on retention is permanent and cannot be undone.
Data Storage and Retention Period in the Security & Fallback Settings panel.
Storage tiers
Each agent uses one storage tier. Every tier has an info icon that explains what it keeps.
Everything (default) stores all call data, including transcripts, recordings, and PII.
Everything except PII stores full call data with the PII categories you select removed. See Redacting personal information.
Basic Attributes Only stores metadata only. No transcripts, recordings, or analytics content are kept.
Hover over a tier’s info icon to see what it keeps.
Retention period
The retention period applies to every tier. Under Retention Period, the options are Keep forever (default), 1 day, 3 days, 7 days, 30 days (1 month), 60 days (2 months), 90 days (3 months), 180 days (6 months), 365 days (1 year), and 730 days (2 years).
When a call’s retention period passes, its stored data is permanently deleted. Basic metadata, such as call ID, timestamp, duration, and status, is kept for reporting. Expired data is removed once a day. When Retention Period is set to Keep forever, call data is not deleted.
Retention Period options.
Setting it up
In the Security & Fallback Settings panel, go to Data Storage.
Select a storage tier.
If you selected Everything except PII, choose the PII categories to redact.
Under Retention Period, choose a retention period.
Click Save.
Changing Data Storage does not change other settings in the panel, such as safety guardrails or the fallback voice.
Data Storage set to Everything except PII.
Per-call override
The tier set on the agent is the default for its calls. Through the API, you can move an individual call to a more restrictive tier, for example from Everything to Basic Attributes Only. You cannot move a call to a less restrictive tier than the agent’s setting. Call history shows the tier and retention applied to each call.
Redacting personal information
PII redaction removes personal information from the agent’s call transcripts, stored recording data, and analytics, while keeping the rest of the conversation. It is available with the Everything except PII storage tier.
The info icon for Everything except PII.
How redaction works
Redaction runs in the background after the call ends, so it adds no delay to the live call. Each piece of detected information is replaced with a labeled placeholder, such as [REDACTED:SSN] or [REDACTED:PHONE_NUMBER], so reviewers can see what type of data was removed.
Redaction follows these rules:
Only the categories you turn on are redacted. Categories you leave off stay visible in the transcript.
Once redaction is on, the original unredacted data is not stored.
Turning on redaction
In the Security & Fallback Settings panel, under Data Storage, select Everything except PII. The Personal Info Redaction (PII) card appears at the top of the panel.
Click Set Up on the card.
The Personal Info Redaction (PII) card with the Set Up button.
Select the categories you want to redact.
PII categories you can select for redaction.
Click Save. Redaction applies to calls after you turn it on. Existing call history is not redacted.
Categories you can redact
The PII categories you can redact.
Category | Data types |
Identity information | Person Name, Date Of Birth |
Contact information | Address, Email, Phone Number |
Government identifiers | SSN, Passport, Driver License |
Financial information | Credit Card, Bank Account |
Security credentials | Password, Pin |
Health information | Medical Id |