Article summary
Viber added ChatGPT chat, @ChatGPT, Image Remix, summaries, polishing, and translation. Review each feature, its privacy boundary, and a safe team checklist.
Rakuten Viber announced ChatGPT features inside Viber on July 7, 2026: a dedicated AI chat, @ChatGPT, Image Remix, chat and link summaries, message polishing, and translation. These are on-demand content tools. They do not represent background access to every conversation and do not automatically create a Viber user activity profile.
For an international team, the value is less app switching and faster multilingual content handling. The risk appears when staff submit customer-sensitive information, internal documents, or unverified AI output directly into business operations.
Usage rule: First decide whether the content should be given to AI at all. For personal data, orders, account security, or legal commitments, de-identify by default and require human review.
What do the six feature groups do?
According to Viber’s official announcement, updating to the latest version provides several ChatGPT experiences. Image Remix requires signing in to or creating a ChatGPT account; at announcement, Viber said the other tools did not require separate registration.
| Feature | Typical input | Output | Useful for |
|---|---|---|---|
| Dedicated ChatGPT chat | Question or idea | Generated response | Public knowledge and drafts |
| @ChatGPT | Mention in a group | Group response | Shared discussion |
| Image Remix | Image and prompt | Edited image | Creative exploration |
| Chat summary | Relevant recent messages | Private summary | Catching up |
| Link summary | Shared article or document | Key points | Initial information screening |
| Polish or translate | Draft text | Rewrite or translation | Cross-language communication |
What content is sent to OpenAI?
Viber says core communication between Viber users remains protected by end-to-end encryption and that ChatGPT features are invoked on demand. For AI features without a connected ChatGPT account, the selected text, image, or request—and recent messages needed when ChatGPT is mentioned—are sent for processing, not the user’s account data or unrelated conversation.
If a user connects a ChatGPT account, related activity in Viber is linked to that account and handled under OpenAI’s standard retention policies. Viber says other AI-feature data is retained for a limited period and is not used to train OpenAI models, while content may be reviewed for safety.
| Situation | Official boundary | Team control |
|---|---|---|
| Ordinary Viber conversation | Core communication is end-to-end encrypted | Still apply internal data rules |
| User invokes AI | Selected request is sent to OpenAI | De-identify and minimize context first |
| ChatGPT account connected | Activity connects to the account and standard retention | Decide whether a work account is appropriate |
| @ChatGPT in a group | Response is visible in the group | Do not submit information that should remain restricted |
Useful applications for a content team
Polishing and translation can support approved drafts without sensitive personal data. Link summaries can help decide whether a public source deserves full review. Image Remix can explore creative directions, but trademark, likeness, misleading presentation, and platform-policy checks remain necessary before publication.
A chat summary may help someone catch up in a busy group, but it should not become the only record for a complaint, contract, or payment dispute. Material decisions require the original message and the system of record.
| Task | AI draft | Human review before use |
|---|---|---|
| Multilingual support reply | Translation and tone | Facts, local language, commitment |
| Event creative | Image Remix options | Rights, brand, people, and text |
| Group follow-up | Summary and actions | Original message, owner, due date |
| Link research | Summary of claims | Open source, verify date and evidence |
Data that should not go directly into chat AI
- Unredacted phone numbers, names, orders, and payment information.
- Passwords, verification codes, recovery codes, private keys, and internal tokens.
- Unpublished contracts, legal advice, or security-incident details.
- Images, documents, or conversation logs without confirmed permission.
- Sensitive data used to decide employment, credit, or other significant rights.
When staff are uncertain, test with a fictional example, field schema, or published material instead of real customer data.
Checks before and after invoking AI
| Point | Question | If it fails |
|---|---|---|
| Before invocation | Does the input contain personal, payment, or security data? | Redact, reduce scope, or stop |
| After generation | Do facts, figures, and commitments have evidence? | Return to official material |
| Before sending | Are language, tone, and recipient correct? | Edit and preview again |
| After sending | Was the final version stored in the formal ticket? | Save the record instead of leaving it only in chat |
If AI mistranslates a customer name, invents a refund promise, or omits a material objection from a summary, log a quality incident rather than silently correcting the sentence. Reviewing incidents by type helps the team decide which tasks can continue, which need stronger approval, and which should be disabled.
ChatGPT in Viber vs Viber Activity Check
| Dimension | ChatGPT in Viber | Viber Activity Check |
|---|---|---|
| Purpose | Questions, creation, summary, translation | Return supported activity fields |
| Input | User-selected prompt, image, or message | TXT with one phone number per line |
| Output | Generated content | Excel: phone number, active days, mid, offline time |
| Proves permission? | No | No |
| Predicts purchase intent? | Not reliably | No |
For a lawfully sourced phone-number list, select Registration, Activity, or another documented task through the AppShai Viber Number Checker. The AI launch did not change those fields and should not be described as ChatGPT identifying active Viber users.
A safe multilingual support workflow
- Copy a case summary without a name, number, or order ID from the ticket system.
- Use translation or polishing to create a candidate response.
- Return to the ticket to verify product facts, prices, and policy.
- Have a capable speaker sample high-risk target-language messages.
- After sending in Viber, save the final response to the formal ticket.
- Escalate complaints, refunds, and legal commitments to a human.
This assigns language assistance to AI without turning chat into the customer master-data store.
Risk matrix before team enablement
| Risk | How it happens | Control |
|---|---|---|
| Sensitive-data exposure | Paste a customer chat | De-identification template and access rules |
| Incorrect response | Treat generation as fact | Source verification and approval |
| Group overexposure | @AI includes restricted context | Approved groups and member training |
| Account mixing | Personal ChatGPT account handles work | Explicit work-account policy |
| Creative infringement | Publish Remix output without review | Rights and brand review |
Deciding whether to enable a feature
For each feature, document purpose, data category, permitted staff, human review, storage location, and stop condition. Pilot on public, low-risk content and compare time saved, rework, errors, and employee experience. Do not invent an efficiency percentage when reliable measurement is unavailable.
If a feature requires data the team is unwilling to give a third-party processor in order to create value, do not enable it. Convenience is not a reason to abandon the data boundary.
