Is Intercom Fin safe for customer conversations? Its model providers are bound by zero-retention, no-training terms, but Intercom’s own Terms, since 8 September 2026, let it use customer data for its own AI model training. Intercom says customers can ask to opt out.
Plain-English risk rating: 2 of 5
Intercom Fin is the most-documented AI customer-service agent in this database from a privacy posture perspective. Intercom has published explicit commitments on its blog and in its Additional Product Terms, and those commitments are about its SUPPLIERS: third-party LLM providers operate under zero-data-retention arrangements and are contractually prohibited from training on conversation data, and the Fin AI Engine is designed with retrieval-and-validation architecture rather than training-on-conversations. Intercom's own Terms of Service, since 8 September 2026, give it the right to use customer data “to provide and improve our Services, including for Intercom AI model training” — see the Training opt-out section below, updated 3 October 2026. Intercom holds SOC 2 Type II certification and supports HIPAA via BAA for qualifying customers, with EU Data Residency available.
The risk picture is dominated less by Fin-specific concerns and more by the broader category challenge for any AI customer-service agent: prompt injection through customer-submitted content, hallucinated responses that could create legal exposure for the merchant, and the operational discipline required to keep Fin's knowledge sources accurate.
Recommended for
- Sole proprietor: Fin is overkill for sole proprietors with low support volume. Use Intercom Inbox without Fin if you're below ~30 tickets per month.
- Small team (2-10 people): Fin at $0.99/resolution makes sense if your support team is genuinely overloaded. The privacy posture is strong enough that the deliberate decision is feature-value, not safety.
- Regulated industry: Fin with BAA, EU Data Residency where applicable, and strict configuration of which Custom Data Attributes Fin can access. Healthcare and finance customers should restrict Fin's knowledge sources to non-PHI/non-financial content.
- The honest answer for most 1-10 employee businesses doing customer support at scale: Fin is one of the best-engineered AI customer-service products as of mid-2026 from a privacy-design perspective. The cost model ($0.99 per autonomous resolution) aligns vendor incentive with actual value delivered. Configure carefully, audit responses regularly, and treat Fin as a tier-1 agent rather than a full replacement.
Data retention default
- Customer conversation content: standard Intercom retention applies per workspace configuration
- LLM provider retention: zero — contractually enforced across providers
- Knowledge sources (articles, content used to train Fin's retrieval): retained while integrated
- Fin generates an audit log of each conversation including which sources were referenced
- Intercom’s revised Data Processing Agreement, effective 4 November 2026, says customer personal data is deleted from its back-up systems 14 days after it is deleted from the live system (section 10), and that Intercom and its sub-processors are subject to laws in the United States and Australia that may require disclosure of, or access to, customer personal data (section 12). It adds terms for the Australian Privacy Principles, including APP 8 on cross-border disclosure. It says nothing about AI or model training, and by default all customer data is stored in the USA unless you use Intercom’s EU or Australian regional hosting (Schedule 2). Read at source 23 October 2026.
Training opt-out
UPDATED 3 OCTOBER 2026 — INTERCOM’S TERMS NOW SAY IT CAN TRAIN ITS AI ON YOUR DATA. Intercom’s Terms of Service, effective 8 September 2026, section 3.2: you grant Intercom the right to use Customer Data “in order to provide and improve our Services, including for Intercom AI model training.” The previous version (9 April 2025) stopped at “provide and improve our Services”. We found no opt-out in the Terms. Intercom’s help article for legal and security teams says its models are “trained using anonymized Customer Data” and that “Customers can request to be opted-out of their Customer Data being used for model training” (read 23 October 2026), so an opt-out exists on request, outside the contract. Customer Data includes your customers’ support conversations. Intercom’s separate commitment that its third-party AI providers may not train on your data still stands (Additional Product Terms, 18 March 2026). Our 21 September correction below said we could not find Intercom’s own position. It was already in the Terms by then, and we missed it. See the corrections page.
CORRECTED 21 SEPTEMBER 2026. This profile previously stated NO TRAINING ON CUSTOMER DATA BY DEFAULT across all Fin tiers. That claim is withdrawn. The quote offered in support of it binds Intercom's suppliers, not Intercom. See the corrections page.
What Intercom's documents actually say. The Additional Product Terms state: "Intercom contractually restricts Third Party AI Providers from using Customer Data for training or otherwise improving Third Party AI Provider's services." The privacy blog this profile cites puts it the same way: "all third-party LLM providers are contractually prohibited from using conversation data to train or improve their models", and "None of our third-party LLM providers store any conversation data." Both commitments bind the THIRD-PARTY model providers, and on their own terms they are strong ones.
What we could not find. Across three Intercom documents read at source on 21 September 2026 – the Additional Product Terms, the Security at Fin help page, and the privacy blog post this profile cites – there is no equivalent commitment that INTERCOM ITSELF will not use Customer Data to train its own models or the Fin AI Engine. The blog does state that "Customer data is only temporarily used to generate responses before being deleted", which speaks to retention rather than to training.
What this means, and what it does not. We have established that our own first-party claim was unsourced. We have NOT established that Intercom trains on your conversations, and we are not saying that it does. A buyer who needs a first-party no-training commitment should ask Intercom for one in writing rather than infer it from the supplier terms. This is general information, not legal advice. (Superseded 3 October 2026: Intercom’s Terms now expressly permit AI model training on Customer Data — see above.)
Zero Data Retention availability
- Default across tiers: Zero-retention API arrangements with LLM providers (OpenAI primarily; Anthropic for some features)
- AI processing for Australian workspaces happens in the US. Updated 23 October 2026: Intercom’s pages now disagree on whether that will change. Its Copilot FAQ (1 September 2026) says regional data hosting for these workspaces “is coming soon”; its earlier article (14 May 2026), quoted below, says it is not planned. Corrected 26 September 2026; this previously said AU workspace data processing was “migrating to AU per Intercom roadmap”, which is the opposite of what Intercom publishes. Intercom states it is making “Fin AI Agent, Copilot, and inbox AI features available to workspaces hosted in Australia” but that “the hosting will remain in the US”, and that “Regional data hosting for these workspaces is currently not planned.” Australian customers must accept Intercom’s Additional Product Terms to use those features. EU workspaces are treated differently: Intercom states EU customers “can now access Fin AI Agent, Copilot, and inbox AI features with data processed within the European Union”. What this means for a New Zealand business: choosing Australian hosting for regional proximity does not keep Fin’s AI processing in the region. If that matters to you, ask Intercom in writing. Read at source 26 September 2026.
Plan tiers and pricing (as of early 2026)
| Tier | Price (USD) | Fin pricing | Suitable for |
|---|---|---|---|
| Essential | $39/seat/month | $0.99/resolution add-on | Small teams |
| Advanced | $99/seat/month | $0.99/resolution included | Growing teams |
| Expert | $139/seat/month | $0.99/resolution included | Larger teams needing SLAs |
| Proactive Support Plus | Add-on | N/A | Outbound messaging |
Plan detail re-checked against Intercom own pricing page on 3 September 2026, and every seat price in the table above was wrong. The current figures are Essential $29, Advanced $85 and Expert $132 per seat per month, against the $39, $99 and $139 shown above. Fin pricing is now quoted as from $0.99 per Fin OUTCOME rather than per resolution, and the distinction this profile drew between Essential having Fin as a paid add-on and the higher plans having it included is no longer right: Intercom includes the Fin AI Agent on all three plans at no extra seat cost, with usage charged separately on every one of them. There is also a standalone Fin option requiring no seats for teams using another helpdesk, which this profile did not carry. None of this changed the training position at the time. The training position HAS since changed – see the correction of 21 September 2026 in the Training opt-out section above, which withdraws the no-training claim this profile previously made about Intercom itself.
Fin Voice (the voice AI agent) has separate fair-use limits per the Additional Product Terms (7-minute average call length, 100 concurrent calls before overage charges apply).
Jurisdiction
- Primary processor: Intercom Inc., San Francisco, California, USA (with Dublin, Ireland EU entity)
- Cloud infrastructure: AWS
- Third-party AI subprocessors: OpenAI (primary), Anthropic (some features) — all under zero-retention and no-training contractual terms
- SOC 2 Type II, ISO 27001 certified
- HIPAA available with BAA at qualifying tiers
- Regional hosting: US, EU or Australia — but read the Australian conditions before planning around them. Corrected 26 September 2026; this previously said “AU Data Residency on roadmap”. Australian hosting exists, and Intercom states it is “only available to new workspaces on the Advanced and Expert plans”, is “not available through the Early Stage program”, is “available to customers on a contract”, and that “Historical conversation data cannot be migrated to the new workspace”. Intercom also states that “Emails (inbound and outbound) will go through the US due to infrastructure limitations”. Read at source 26 September 2026.
Breach history (public incidents)
No major direct breach of Intercom infrastructure publicly disclosed as of May 2026.
Intercom has published a transparency document acknowledging that "over the past year, we've observed very few successful attempts at bypassing the safeguards put in place to protect our AI-powered features. None of these attempts constituted a reputational threat to our customers, and all incidents were mitigated in a timely fashion." This level of public acknowledgement of attempted bypasses is uncommon and reflects a relatively mature security posture.
Category-level risk: As with any AI customer-service agent, prompt injection via customer-submitted content (a customer typing instructions intended to manipulate Fin) is the dominant attack class. Intercom's retrieval-and-validation architecture reduces this risk compared with pure-LLM-response systems, but does not eliminate it.
What this means in plain English for SMB owners
Three honest takeaways:
- Intercom Fin's privacy and security documentation is among the best in this database. The combination of zero-retention contracts with LLM providers, SOC 2 Type II, HIPAA availability, and EU residency is a defensible enterprise-grade posture at SMB pricing – though the no-training commitments bind Intercom's suppliers rather than Intercom itself, corrected 21 September 2026.
- The $0.99-per-resolution pricing model genuinely aligns Intercom's incentive with delivered value. Compare this with seat-based AI pricing where the vendor benefits from heavy usage regardless of resolution quality. Audit your monthly Fin resolution rate against escalations to track whether Fin is actually resolving or merely punting to humans.
- Operational discipline matters more than vendor choice for AI customer service. Whatever AI agent you use, the determinants of safety and quality are (a) how accurate your knowledge sources are, (b) which Custom Data Attributes you let the AI access, and (c) how clearly you have configured escalation rules for sensitive topics (refunds, cancellations, regulated data). Fin makes these configurable; the configuration is the work.
Sources
- Intercom Additional Product Terms: https://www.intercom.com/legal/terms-and-policies/additional-product-terms (verified 2026-05-24)
- Intercom blog: How Intercom ensures data privacy and safety in the age of AI (February 2025)
- Intercom Help: The Fin AI Engine documentation
- Intercom Help: Fin AI Agent FAQs
- Fin.ai: HIPAA & GDPR Compliant AI Agents guide (April 2026)
- My AskAI: Intercom Fin AI Agent Complete Guide (March 2026)
Related on AI Leakage
- Compare all 29 AI tools in the risk directory — see how Intercom Fin stacks up against the rest, tier by tier.
- Take the 5-minute “Am I Leaking?” check — a personalised view of your business’s AI exposure.
- Check a prompt before you paste it — our free Data-Safe Prompt Rewriter.
- Shadow AI vs AI leakage — why even approved tools like Intercom Fin can leak data.
- Get plain-English AI Leakage Alerts — we email you when an AI tool you use changes its data policy or has an incident.
- Get the free AI Acceptable Use Policy template — a plain-English policy with the tool-by-tool risk guide built in.
Compare similar tools: Intercom Fin is one of 4 tools in AI in customer service and CRM compared, which sets out each one’s rating and what it does with your data, side by side.