OpenAI tests 'o' assistant: always-on ChatGPT with email hints

References to an unannounced always-on ChatGPT assistant called "o" briefly appeared online this week, suggesting that OpenAI is testing a new consumer-facing feature that could handle tasks even when the user is not actively chatting. The discovery, shared on X by a user named Jake and reported by BleepingComputer, shows the phrase "o, your always-on assistant" listed among the benefits of the $100 ChatGPT Pro plan, a subscription tier that already includes increased usage limits, maximum memory, 100 GB of file storage, and early access to upcoming features. OpenAI has not officially confirmed the project, but the listing appeared live for a short period before being taken down.

More intriguing is a separate dynamic configuration file that was also spotted online. This configuration contained a reference to a display name of "o" and, crucially, an email suffix of "-o." While the exact purpose of this suffix remains unclear, it strongly hints that the assistant could adopt an email-based identity or possess some form of email-handling capability. An always-on assistant with email access would, in theory, be able to draft replies, summarise threads, or manage correspondence in the background, stepping beyond the session-based interactions that current ChatGPT users experience. BleepingComputer cautions that we do not yet know exactly what OpenAI has planned, and the company has neither confirmed nor denied the feature.

The timing of these leaks is notable. OpenAI has scheduled its DevDay 2026 event for September 29 in San Francisco, where it is expected to announce several new tools, host technical sessions, and provide demos of upcoming work. The appearance of "o" references just days before DevDay suggests that the assistant could be one of the reveals onstage. The same source report mentions a separate internal system code-named "Aeon," which appears related to custom agents for ChatGPT workspaces, but the "o" assistant is described as a distinct consumer experience rather than an enterprise tool.

For users and businesses that rely on ChatGPT for productivity, an always-on assistant with email integration represents a significant shift in how AI tools might blend with daily workflows. Rather than opening a dedicated chat interface, an always-on assistant could respond to notifications, compose messages, or fetch information from connected accounts without explicit prompting. However, the concept also introduces practical security and privacy questions, particularly if the assistant is granted ongoing access to email inboxes. Granting such persistent access to any service requires careful scrutiny of permissions, data handling, and the assistant's ability to act autonomously, which OpenAI will need to address transparently if the feature launches.

The brief appearance of these references underscores a pattern in which unreleased features surface through web configurations or benefit listings, often giving the public an early glimpse before official announcements. OpenAIs silence on the matter is typical for features still in testing, and until more details emerge at DevDay or through an official blog post, the full scope of "o" remains under wraps. Website owners and IT teams who may one day connect an AI assistant to their business email or hosting dashboards should recognise that any third-party tool granted persistent account access adds a new layer of risk. For those managing WordPress sites, using a hosting platform that already bakes in security monitoring and access controls, such as the managed protections within AEU Hosting, helps ensure that even if an email or assistant account were compromised, the hosting environment and customer data would not become an easy next target.

As AI assistants grow more autonomous, the boundary between a helpful tool and an over-privileged application narrows. Until OpenAI clarifies how "o" would authenticate, what data it stores, and whether users can revoke access instantly, the safest approach for anyone who might adopt such a service is to treat its permissions with the same caution applied to any application that can read, write, and send on their behalf.