WhatsApp Tests Third-Party AI Agents in Android Beta

WhatsApp has begun testing a feature that will let users connect AI agents built by third-party companies to the app and chat with them in separate, dedicated conversations. The addition was spotted in the Android beta version of the app and is currently rolling out to a small group of testers in select countries.

How the new section works

The change was first reported by WABetaInfo, a site that closely tracks the code of WhatsApp test builds. A new Agents tab appears in the app's settings, where users can add a third-party service, give it a name, and assign it a profile picture. Once configured, the agent shows up in the chat list as a separate, clearly marked entry.

On the technical side, the process resembles connecting an integration through an API. WhatsApp generates a unique key that the agent's creator must enter on their own platform hosting the service in order to link it to a specific chat. Only after that step does the agent start responding in the app as if it were just another contact.

Unlike simple bots built around rigid commands, the new agents are meant to understand more complex requests and carry out specific tasks on the user's behalf, rather than just returning canned answers from a list. That makes them similar to the coding agents and shopping assistants that other platforms have been testing in recent months.

What happens to privacy

The key caveat concerns encryption. Standard WhatsApp conversations are protected end-to-end, but messages sent to third-party agents will pass through a service managed by Meta on behalf of the agent's creator, meaning the agent's developer may be able to access them. The agent itself, however, will not see the user's other chats, contact list, or media outside the dedicated conversation with it.

Users will be able to change an agent's name and picture at any time, or disconnect it from their account by revoking the API key. The chat history from the conversation is set to remain saved in the app even after the service is disconnected.

A reversal after the November ban

The test carries extra weight given a decision Meta made less than a year ago. In November 2025, the company announced that starting January 15, 2026, it would ban competing chatbots such as ChatGPT or Copilot from WhatsApp, citing new platform terms of service. The ban mainly targeted consumer-facing tools competing directly with Meta AI, with an exception for customer-service bots used by businesses on WhatsApp Business.

The new agents feature, however, operates under different rules than those blocked integrations. All traffic runs through infrastructure controlled by Meta, and access to the platform requires registration via an API key generated by the app itself. In other words, Meta is not reopening WhatsApp to rivals, it is building its own controlled gateway through which third-party agent developers can offer their services on Meta's terms.

What it means for developers and businesses

For AI agent developers, opening another distribution channel inside an app with more than two billion users worldwide is a potentially major opportunity, especially in shopping assistants, booking tools, or handling official/administrative matters. At the same time, the lack of end-to-end encryption and the requirement to route through Meta's infrastructure mean Meta decides the business and technical terms under which such an integration is possible.

WhatsApp has not yet officially confirmed a timeline for a wider rollout of the feature or named the first partners it might work with. Testing on a limited group of Android users is a typical way for Meta to validate new mechanisms before they reach the public beta and, later, the stable version of the app.