> ## Content Index > Fetch the complete content index at: https://bytevyte.com/llms.txt > Use this file to discover other available public pages before exploring further. # iPhone Beta Testers Can Now Add WhatsApp Third-Party AI Agents, But Chats Stay Unencrypted [Update] - URL: https://bytevyte.com/iphone-beta-testers-can-now-add-whatsapp-third-party-ai-agents-but-chats-stay-unencrypted-update/ - Published: 2026-09-25T18:53:28.000Z - Updated: 2026-09-25T18:53:28.000Z - Description: WhatsApp third-party AI agents reach iPhone beta 26.37.10.70 with a five-agent cap, API-key setup and no end-to-end encryption on agent chats. - Author: Bytevyte Editorial - Tags: quick-beats The rollout of WhatsApp third-party AI agents to iPhone testers has begun, with the feature appearing in the iOS beta build **26.37.10.70** on TestFlight. Each account can hold up to five agents, and access remains limited to a small number of users in selected countries. The same capability surfaced on Android earlier this month, as we previously reported. Setup lives inside WhatsApp's settings. A user gives an agent a name and a profile photo, then receives an API key generated by WhatsApp and enters it into whichever outside service hosts the bot. After linking, the agent appears as its own one-to-one thread that looks like any other conversation. The agent does not have to be an AI system; a conventional bot connects through the same key. ## How Agent Chats Differ From Regular Messages The distance between a normal WhatsApp conversation and one with an agent comes down to who can read the messages and what the thread supports. | Aspect | Standard WhatsApp chat | Third-party agent chat | | ------------------------- | ---------------------- | ---------------------------------------------------- | | Encryption | End-to-end | Not end-to-end | | Who can read messages | Only the participants | The agent developer, through a Meta delivery service | | Disappearing messages | Supported | Not supported | | Group and community chats | Supported | One-to-one threads only | | Agents per account | Not applicable | Five, a limit WhatsApp may change | Two of those rows carry the practical weight. Because agent messages pass through a service that runs on the developer's behalf, the developer can read them, and a thread with no disappearing-message option is a thread with no built-in expiry. ## The Trade-Off Behind WhatsApp Third-Party AI Agents End-to-end encryption works because WhatsApp cannot read message contents. An agent hosted on an outside platform cannot fit that model, since the external server has to receive the text in order to answer it. WhatsApp routes each message through a Meta service that handles delivery on the developer's behalf, and the company therefore places agent conversations outside its end-to-end encryption guarantee. The missing disappearing messages option follows from the same design: a thread whose contents rest on a third-party server cannot offer the same retention behaviour as a chat between two phones. WhatsApp does contain the exposure. An agent reads only what a user types into its own thread and holds no access to other conversations, the contact list, or media stored on the device. That containment matters for anyone deciding how much personal detail to type into a bot. It does not change the underlying fact that agent messages leave the encrypted envelope, and it says nothing about how a developer stores or reuses the text afterward. The API key is where the trust decision actually happens. WhatsApp generates it, and the user carries it to the service that hosts the bot, which means the linking step grants that developer a channel into the account before a single message is exchanged. Everything after that runs on terms set by the bot's operator, not by WhatsApp, and the containment WhatsApp does enforce covers scope rather than confidentiality. The design choice worth noting is placement. Agent threads sit in the same chat list as family, friends and colleagues, separated only by a name and photo the user picked. That makes agents easy to reach and easy to mistake for people, particularly when a bot is built to reply conversationally. ## What Users And Developers Actually Get For users, the appeal is convenience. Tasks such as planning a day, organising an activity, or drafting a message can be handled where conversations already happen, and each agent gets a name and profile photo that make it read like a contact. For developers, the API key route removes the need to build a separate app or pull users off WhatsApp to reach them, and it avoids maintaining a second identity for the same bot. Nothing in the current test suggests WhatsApp plans to fold its own model into these threads. The agents are built and hosted elsewhere, which makes this a distribution channel for outside bots rather than a first-party assistant. The limits are as concrete as the capabilities. Agents work only in one-to-one threads, so they cannot join group chats or communities at this stage. The five-agent ceiling is a beta setting that WhatsApp may adjust. Availability stays narrow: a limited set of users in select countries, and WhatsApp has not signalled how quickly that group will grow or when a wider release might follow. Reading the five-agent cap as a settled product decision may overstate it. WhatsApp has said the number could change, which suggests the ceiling exists to keep the test measurable rather than to define a permanent limit. The one-to-one restriction is firmer: an agent cannot be dropped into a family thread or a work group and answer on everyone's behalf. The iOS build mirrors the Android test in architecture and cap, which points to WhatsApp validating one setup across platforms instead of shipping different rules per operating system. A non-technical user still needs a bot that already hands out a key, so early adoption will lean on services that have prepared for the integration. Anyone weighing the feature should separate the two questions it raises. The first is whether the bot is useful enough to justify stepping outside an encrypted thread. The second is what the operator of that bot does with the text afterward, a matter WhatsApp's terms do not settle. Users who keep sensitive conversations in ordinary chats and reserve agent threads for low-stakes tasks stay on the safe side of a boundary the beta does not yet let them see through. ## Why this matters WhatsApp is testing whether outside bots can live inside its interface without inheriting its privacy promises. For users, the practical rule is to treat an agent thread like a web form rather than a private chat: share what the task requires and nothing more. If the beta widens, the five-agent cap and the encryption boundary are the two settings worth watching, because together they define how much of WhatsApp's messaging stays private and how much becomes an integration surface. *See our earlier coverage:* [*Third-Party AI Agents on WhatsApp Reach Android Beta Testers With an Encryption Catch*](https://bytevyte.com/third-party-ai-agents-on-whatsapp-reach-android-beta-testers-with-an-encryption-catch/) Photo by [JF Martin](https://unsplash.com/@numericcitizen?utm%5Fsource=bytevyte&utm%5Fmedium=referral) on [Unsplash](https://unsplash.com/?utm%5Fsource=bytevyte&utm%5Fmedium=referral) ## Related Articles - [Third-Party AI Agents on WhatsApp Reach Android Beta Testers With an Encryption Catch](https://bytevyte.com/third-party-ai-agents-on-whatsapp-reach-android-beta-testers-with-an-encryption-catch/) - [Apple Prepares iOS 27 with Dedicated Siri Chatbot and Visual Intelligence Features](https://bytevyte.com/apple-prepares-ios-27-with-dedicated-siri-chatbot-and-visual-intelligence-features/) - [iPhone Users Can Now Test iOS 27 Public Beta With Revamped Siri AI Starting This Week](https://bytevyte.com/iphone-users-can-now-test-ios-27-public-beta-with-revamped-siri-ai-starting-this-week/) ✔Human Verified --- *Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.*