OpenAI’s recent DevDay event revealed ChatGPT’s evolution toward an “everything app” with new platform capabilities. A key missing component, user-to-user direct messaging, now appears imminent following discoveries in a beta version of its Android application.

The announcements from the developer conference centered on transforming ChatGPT into a platform that can run applications directly within chats, a move that positions the service to function more like an operating system. This evolution aligns with potential plans for a first-party ChatGPT hardware device, which reports suggest might launch in late 2026 at the earliest. Combined with previously announced support for shopping functionalities, these changes push ChatGPT toward the model of an “everything app” with artificial intelligence at its functional core. A central feature of many such comprehensive platforms is a native communication tool for user-to-user interaction, a function that has been absent from ChatGPT.

Evidence that OpenAI is developing such a feature first surfaced in early October. The discovery was made by Tibor Blaho, an individual who frequently shares information about unreleased AI products online. Within a beta version of the ChatGPT application for the Android operating system, Blaho found code referencing direct messaging (DM) functionality. The feature was identified by the codenames “Calpico” and “Calpico Rooms.” This naming convention and function may be similar to the direct messaging capabilities that OpenAI previously developed for its Sora application on the iPhone, indicating a potential template for the new system.

“Direct Messages” has now been added to the ChatGPT Android app (version 1.2025.273 beta) – codename “Calpico” / “Calpico Rooms”

Direct messaging was first introduced yesterday in the new Sora 2 iOS app, and now code references in the ChatGPT Android app show it includes direct… pic.twitter.com/e2kzWyu4N9

— Tibor Blaho (@btibor91) October 1, 2025

Further evidence emerged on Saturday, when Blaho provided an update from a different Android beta version of the ChatGPT app. He posted a screenshot of the user interface that included a distinct icon designated for direct messaging. These successive findings in pre-release software builds strongly indicate that a system for user-to-user chat is not only in development but is progressing toward an eventual public release. This development raises a critical question regarding the feature’s technical implementation, specifically whether it will incorporate end-to-end encryption (E2EE) to match the security standards of established messaging services like iMessage, WhatsApp, and Signal.

New ChatGPT Android app beta (1.2025.280) adds more references related to “Direct Messages” / “Group Chats”

– turn on/off assistant auto-response (respond automatically/when mentioned)
– rename the assistant and update assistant personality
– block accounts
– brainstorm ideas,… https://t.co/ILJeRJ0EBd

— Tibor Blaho (@btibor91) October 11, 2025

Meta has already integrated its Meta AI assistant across its suite of applications, including the E2EE-enabled platforms WhatsApp and Facebook Messenger. A similar strategy by OpenAI, building a DM feature within its core chatbot app, would be a logical extension of its platform. The utility of such a feature is apparent for anyone who has used the existing “Share” function for ChatGPT conversations, as a native DM system could provide a more integrated and secure method for collaboration and information sharing directly within the application’s environment, removing the complexities of managing external shared links.

Blaho’s initial findings in October included specific notes on privacy. He stated that while users could engage with ChatGPT in group chats, their personal memories stored within the platform would not be accessible to the AI or other participants in these direct messaging rooms. The system is also being designed to send users notifications for new messages and for events within a chat room, such as a new participant joining or an existing one leaving. Over the weekend, he expanded on the feature’s intended functions. The DM system is designed to allow users to “ChatGPT together,” a collaborative experience with a range of capabilities.

• Collaborative Tools: Users can brainstorm ideas, create plans, ask questions, and conduct searches as a group.
• Content Creation: The feature will support joint image creation within the chat interface.
• Assistant Customization: Participants will have the ability to rename their assistants and update their personalities for the context of the group chat.
• Chat Management: The system includes options to invite others to group chats and to block participants.
• Automation: An auto-response option is also part of the planned functionality.

The question of how OpenAI will address user privacy for this direct messaging feature is complex. OpenAI CEO Sam Altman has spoken about the importance of ensuring conversations with AI remain private, proposing a concept of “AI privilege” that would offer protections analogous to doctor-patient or attorney-client confidentiality. This level of privacy becomes particularly critical in the context of a future personal assistant device, which would likely handle highly sensitive user data. For such a device to be trusted, its underlying communication systems would require robust security measures.

However, OpenAI’s current data handling practices present challenges to implementing comprehensive privacy protections like E2EE. The company utilizes user chat data to train its models, a process that users can opt out of but which is enabled by default. Furthermore, OpenAI employs autonomous systems to scan conversations for potential violations of its usage policies and to monitor for ideation of self-harm. In cases where the automated systems flag content, human review teams may be involved, meaning those chats are not entirely private. These internal monitoring processes are fundamentally incompatible with a system where the service provider cannot access message content.

External legal and technical issues also complicate the privacy landscape. Earlier this year, in a case brought by The New York Times, a court ordered OpenAI to prevent the permanent deletion of user chats that had been marked for deletion. While a recent modification to the order allows the company to now delete some of that preserved data, the case highlights how legal obligations can override user privacy controls. In a separate incident during the summer, a technical issue caused shared ChatGPT conversations, which users intended to be semi-private, to be indexed by Google Search, exposing them publicly. These events underscore existing gaps in the full protection of user data on the platform.

These factors—data use for model training, content monitoring for policy enforcement, legal data preservation orders, and past privacy vulnerabilities—create a difficult environment for implementing E2EE from the feature’s inception. A failure to include this level of security would result in a DM experience that is less secure than many of its direct competitors. If OpenAI’s ambition is for ChatGPT to mature into a full-fledged operating system, securing its core components, including any first-party communication apps, with strong encryption will be a fundamental requirement.

Featured image credit