WhatsApp is testing a built‑in way for users to connect third‑party AI agents directly inside the app, giving each connected service its own conversation. The capability is rolling out to a limited number of Android beta users and — according to multiple reports that cite a WABetaInfo discovery — supports up to five agents per account, uses API keys to link services, and does not apply WhatsApp’s usual end‑to‑end encryption to those chats.
What the beta shows
Across reports from 9to5Mac, iThinkDiff, MobiGyaan, Newsbytes, ShiftDelete and others, the same basic feature details appear:
- WhatsApp’s Agents section (spotted in WhatsApp beta for Android version 2.26.35.3) lets users add and configure compatible third‑party services.
- When a user creates an agent entry, WhatsApp generates an API key the user copies into the external AI platform to complete the link.
- Each connected agent gets a dedicated one‑to‑one chat, with a name and profile photo that appear in the chat list.
- WhatsApp reportedly allows up to five agents per account.
- Agents can only read the messages sent to their dedicated chat; they cannot access other chats, contacts, or media outside that chat, per the reports.
- Agent conversations cannot be added to group chats or communities.
Encryption and developer access — reports disagree on language but agree on outcome
All sources indicate agent chats are not protected by WhatsApp’s standard end‑to‑end encryption. Several reports add context about how the messages flow: WhatsApp routes messages to a Meta service on behalf of the agent’s developer, and the developer running the agent can access conversation content as part of operating the service.
The sites use slightly different wording — for example, iThinkDiff says messages are “secure in transit, but not end‑to‑end encrypted in the traditional sense,” while MobiGyaan and Newsbytes frame the flow as passing through a Meta service that allows developer access — but they consistently report that these agent chats do not carry WhatsApp’s usual E2EE protections.
Why this matters
This would change WhatsApp’s prior stance toward unofficial bot integrations. Several reports note that earlier attempts to link custom AI agents to WhatsApp sometimes led to account suspensions, and that Telegram had become a common alternative because it supported bots. Native agent support would remove that risk and make it easier for users to interact with AI tools inside WhatsApp’s interface.
However, the privacy trade‑offs are material: users who value WhatsApp’s end‑to‑end encryption will be exchanging that guarantee when they communicate with third‑party agents, and developers will be able to read messages sent to their agents.
Practical checklist for users considering agents
- Confirm beta access: the feature is appearing in a limited Android beta (reported build 2.26.35.3). If you don’t see an Agents section, you likely don’t have the test yet.
- Name and identify agents: assign a clear name and profile photo so you can distinguish multiple agents in the chat list.
- Treat agent chats as service conversations: do not send passwords, highly sensitive personal data, or unencrypted financial information to an agent chat because those messages are accessible to the agent’s developer.
- Check the agent developer’s privacy and data‑retention policies before pasting any API key or sending production data to the linked service.
- Use the app’s settings to regenerate API keys if you need to reset a connection or revoke access.
- Keep personal one‑to‑one and group chats for private conversations that require end‑to‑end encryption.
Limitations, rollout and unresolved points
All available reporting traces back to the beta discovery; WhatsApp (Meta) has not provided an official public announcement in these reports. The timeline for a broader rollout remains unclear. Sources agree the initial testing is limited to a small number of Android users and that the company could expand availability later.
Open questions the beta reports do not resolve include whether WhatsApp will require developers to meet specific security or compliance standards to register as agent providers, whether Meta will offer a developer portal or certification process, and how long Meta retains agent chat metadata or message content when routing messages on behalf of developers. Those details are not present in the cited reports.
Evidence comparison and reader guidance
The six reporting outlets in the evidence packet (9to5Mac, iThinkDiff, Newsbytes, MobiGyaan, ShiftDelete and syndicated MSN content) all cite the same initial WABetaInfo discovery. They agree on the feature’s core mechanics: API key linking, a five‑agent limit, individual agent chats, and the absence of end‑to‑end encryption for those chats. Wording around how messages are handled varies slightly, but the substantive claim — developers can access agent chat content and E2EE does not apply — is consistent across sources.
If you rely on WhatsApp for sensitive communications, treat agent chats as third‑party service interactions until (and unless) Meta publishes formal documentation specifying stronger protections. For developers and AI service operators, plan for an API key model, clear user disclosures, and straightforward ways for users to revoke or rotate keys.
Reported sources: WABetaInfo discovery as relayed by 9to5Mac, iThinkDiff, Newsbytes, MobiGyaan and ShiftDelete.