Skip to main content

Safety Guard

An overview of Safety Guard — what it is, the topics it screens for, and how the two-strike system keeps Fin's responses safe

Written by Dawn Perrott

Use this article to understand how Safety Guard screens customer messages, see which topics are always declined, and let Fin handle flagged messages using your own procedures or knowledge content instead of declining them.

What is Safety Guard?

Safety Guard is Fin's built-in shield against sensitive and harmful content. It screens every message automatically, so your customers always get a safe, considered response, and your team only steps in when it counts.

Safety Guard is part of Intercom's AI Usage and Conduct certification (AIUC-1) and is enabled by default for all workspaces using Fin. It applies to all plan types and cannot be disabled.

Note: Safety Guard applies to all Fin workspaces, regardless of plan. Letting Fin handle flagged messages currently applies to Fin AI Agent for Service only.

What Safety Guard screens for

Safety Guard screens for nine topic categories:

  • Self-harm and suicide

  • Violence and threats

  • Hate speech and discrimination

  • Illegal activity

  • Adult and sexual content

  • Dangerous substances

  • Weapons

  • Privacy and personal data misuse

  • Fraud and deception

Topics that are always declined

No setting changes how Fin responds to messages about these topics — Fin always declines them, even when you let Fin handle other flagged messages with your own procedures or knowledge content:

  • Self-harm and suicide

  • Child safety

  • Vulnerable people

  • Abuse, and attempts to misuse Fin

Important: For these topics, Fin always gives the standard reply and offers a teammate, no matter which options you turn on below.

How Safety Guard works

Safety Guard operates on a two-strike system designed to balance customer safety with conversational continuity. When Safety Guard flags a message and no option lets Fin handle it (see Let Fin handle flagged messages below), Fin sends this standard reply and offers a teammate:

Note: “I'm not able to help with that topic, but I can connect you with a member of the team who can. Would you like me to do that?”

Strike 1: first detection

When a customer's message or Fin's draft response is flagged for the first time in a conversation, Safety Guard:

  1. Blocks the flagged content from being sent or processed further.

  2. Delivers a configured safe response to the customer instead.

  3. Logs the detection internally.

The customer can continue the conversation normally afterward. Safety Guard does not end or close the conversation at this stage.

Strike 2: second detection in the same conversation

If a second screening flag is triggered in the same conversation, Safety Guard automatically escalates to your human support team. The conversation is routed away from Fin so a teammate can take over directly. Once a conversation has been escalated after a second strike, Fin does not re-engage in that conversation.

Important: The two-strike count resets at the start of each new conversation. A customer who triggered strikes in a previous conversation starts fresh.

Let Fin handle flagged messages

Why use these options

For some businesses, sensitive questions are part of everyday support. A health provider gets questions about dosage. A legal or tax product gets questions about rules and deadlines. If you already have approved content or a procedure for these questions, two options let Fin use it, so the customer gets help without waiting for a teammate.

The two options

Two toggles in Settings let Fin use your own knowledge content or procedures for flagged messages, instead of declining them. You can turn on one option or both — both are off by default.

Answer from your knowledge: Fin answers select questions from your knowledge when it finds a grounded answer. Otherwise, it gives the standard reply. Fin answers only when your content supports the answer — if your content doesn't cover the question, the customer gets the standard reply.

Route to procedures: When a flagged question matches one of your procedures, Fin runs it instead of declining. Fin only runs a procedure if it matches the question in the normal way — the option doesn't make Fin choose a procedure it wouldn't choose otherwise. The procedure must be live for that customer.

Turn on an option

  1. Go to Settings > AI & Automation > Fin AI Agent and expand Fin's Safety Guard.

  2. Turn on Answer from your knowledge, Route to procedures, or both.

  3. Read the confirmation, then select the checkbox to accept it on behalf of your workspace, and select Accept.

  4. Your option is now active. In the Inbox, a conversation where a message was flagged shows the event Fin's safety guard was triggered — expand it to see the topic and what Fin did.

To turn an option off, turn off its toggle and confirm. Fin then declines these messages again and offers a teammate. You can change these options at any time.

What happens when a message is flagged

With one or both options on, Fin checks the following for a flagged message, in order:

  1. If the message is about a topic that's always declined, Fin gives the standard reply.

  2. If your own escalation rules or guidance hand the conversation over to a teammate, Fin hands it over. The options never stop a handover you've set up.

  3. If Route to procedures is on and a live procedure matches, Fin runs that procedure.

  4. If Answer from your knowledge is on and your content supports an answer, Fin answers.

  5. If none of the above apply, Fin gives the standard reply and offers a teammate.

Topics that are always declined stay declined at any point in the conversation.

Set up your procedures for sensitive topics

  • Make sure the procedure is live. If it's paused, in draft, or limited by an audience rule, Fin can't use it and declines the message. This is the most common reason that Route to procedures seems to do nothing.

  • Check your escalation guidance. If you tell Fin to hand over all symptom reports, for example, those messages go to a teammate and not to your procedure.

  • Send customers to specific content. You can't pick specific articles for flagged messages only. Instead, create a procedure whose reply links to the articles you want — for example, a page that tells customers to see a medical professional — and turn on Route to procedures.

Note: These options currently apply to Fin AI Agent for Service only.

FAQs

Can I disable Safety Guard for my Fin workspace?

You can't disable Safety Guard for your Fin workspace. Safety Guard is enabled by default on all Fin workspaces across all plan types and can't be turned off. It's a core requirement of Intercom's AI Usage and Conduct certification (AIUC-1). If Safety Guard declines questions that your business handles every day, such as medical, legal, or financial questions, let Fin handle them using your own procedures or knowledge content instead — Safety Guard stays on, but Fin can now answer or route those questions rather than declining them.

Does Safety Guard affect all customer conversations, or only sensitive ones?

Safety Guard does not affect all customer conversations — it only activates when a message or Fin's draft response matches one of the nine screened topic categories. Normal conversations that do not touch a screened topic are completely unaffected by Safety Guard.

What happens to a customer's message when Safety Guard flags it?

When Safety Guard flags a customer's message, the flagged content is blocked and not processed further. If you've turned on Route to procedures or Answer from your knowledge and the message qualifies, Fin routes it to a matching procedure or answers it from your content. Otherwise, the customer receives the standard reply with the offer of a teammate. If it was Fin's own draft response that was flagged, that response is blocked before it reaches the customer, and the standard reply is sent in its place.

Can I turn off Safety Guard for one topic only?

You can't turn off Safety Guard for one topic only. Instead, let Fin handle flagged messages with your procedures or content using the options described above. Topics that are always declined stay declined regardless.

Does turning on an option switch off Safety Guard?

No, turning on an option doesn't switch off Safety Guard — it still checks every message. The options only change what Fin does after a message is flagged, and only for topics that aren't always declined.

Can I choose which topics the options apply to?

You can't choose which topics the options apply to. Each option applies to all the topics that can be overridden. Topics that are always declined are never included, no matter which options are on.

I turned on Route to procedures but Fin still declines. Why?

If Route to procedures is on but Fin still declines a flagged message, check that a procedure matching the question is live for that customer and not limited by an audience rule. Also check whether your escalation guidance hands the conversation over to a teammate first — the options never override an existing handover.

Does Fin check my content before it answers a flagged message?

Fin doesn't check your content for appropriateness before answering a flagged message. Fin answers only when your content supports the answer, but it doesn't judge whether the content is appropriate for a sensitive topic — your workspace is responsible for that.

Can I see when Fin used an option?

You can see when Fin used an option in the conversation in the Inbox. When Safety Guard flags a message, the conversation shows the event Fin's safety guard was triggered — expand it to see the topic and what Fin did: routed to a procedure, answered from your content, declined, or escalated (the second flag in the conversation). If your own rules or guidance handed the conversation over to a teammate first, this event doesn't show, and there's no separate report for these events.

💡Tip

Need more help? Get support from our Community Forum
Find answers and get help from Intercom Support and Community Experts

Did this answer your question?

Related Articles