OpenAI ‘o’ Leak: What We Know About ChatGPT’s Always-On Assistant Before DevDay

References to an unannounced OpenAI assistant called “o” have surfaced in ChatGPT, suggesting the company is testing an always-on assistant that may have some form of email functionality.

OpenAI has not announced the product, described its feature set, or confirmed a release date. With DevDay taking place Sept. 29 in San Francisco, the leak raises questions about whether “o” could become a new consumer-facing layer for tasks that continue after a user leaves ChatGPT.

According to BleepingComputer, “o, your always-on assistant” briefly appeared among the benefits shown for a $100 ChatGPT Pro plan.

A separate dynamic configuration contained display_name: "o" alongside an email_suffix: "-o" field.

That second reference is the strongest clue that email could be involved, but it does not establish what “o” would actually do with email. The configuration does not confirm that the assistant can read inboxes, send messages, forward email, or manage an account.

What the OpenAI “o” leak actually shows

The leaked references point to three details: the name “o,” the description “always-on assistant,” and an email-related configuration field.

BleepingComputer reports that “o” appears separate from another internal OpenAI project called Aeon, which has been associated with custom agents for ChatGPT workspaces.

That distinction could place “o” closer to a consumer-facing ChatGPT feature than to another workspace-agent product, although OpenAI has not confirmed that interpretation.

The appearance of “o” alongside ChatGPT Pro benefits also suggests OpenAI has at least tested where the assistant might fit within its subscription lineup. It does not confirm that “o” will launch with Pro, remain exclusive to that plan, or use the $100 price shown when the reference appeared.

The email_suffix: "-o" reference is similarly limited. It could indicate an email identity for the assistant, a routing mechanism, or functionality connected to email workflows. The available evidence does not distinguish among those possibilities.

There is also no public evidence yet for other capabilities that might be associated with an always-on assistant, such as proactive notifications, autonomous web activity, calendar management, or background research.

OpenAI already has technology for long-running agents

Although “o” itself remains unconfirmed, OpenAI already has products that continue to work after a user steps away.

OpenAI’s workspace agents can run on schedules, use connected tools, and handle repeatable workflows. The company says its own sales team uses an agent to combine call notes and account research, and to draft follow-up emails in a representative’s inbox.

That is an existing workspace product, not evidence that “o” will inherit the same capabilities.

OpenAI has also pushed ChatGPT beyond conventional chat through ChatGPT Work, which can work across files and connected services and continue multi-step tasks.

Its newer Agents API provides infrastructure for agents that OpenAI says can run reliably for days while preserving context across longer sessions. OpenAI describes the API as using the managed Codex harness and infrastructure that powers Codex.

Recent products such as GPT-6 Astra have also expanded OpenAI’s agent-focused capabilities across ChatGPT and Work.

Those products make the phrase “always-on assistant” consistent with OpenAI’s broader product direction. They do not establish whether “o” uses workspace agents, ChatGPT Work, the Agents API, or a separate system.

Email may be the biggest unanswered feature

The leaked email suffix creates the clearest unanswered question.

OpenAI already supports email-related workflows through connected tools and workspace agents. Its documentation also describes approval controls that can require permission before an agent sends an email, modifies a spreadsheet, or creates a calendar event.

TechRepublic has previously examined how OpenAI’s agent controls use scoped access and authorization to limit what agents can reach.

None of those controls have been confirmed for “o.”

If the assistant does connect directly to email, the important product details will include whether it can only draft messages or also send them, whether users must approve outbound actions, which inboxes it can access, and whether those permissions can be limited.

The leak does not answer any of those questions yet.

DevDay could provide the next clue

OpenAI DevDay takes place Sept. 29 at Fort Mason in San Francisco. Sam Altman is scheduled to open the event with a keynote at 10 a.m. Pacific time.

OpenAI’s official DevDay schedule does not mention “o,” and the company has not said the assistant will be announced at the event.

BleepingComputer cited DevDay as a possible source of more information, but that remains speculation.

For now, the evidence supports a narrower conclusion: OpenAI has tested references to a product called “o” described as an always-on assistant, and a separate configuration associates that name with an email-related field.

Everything beyond that, including how it works, which ChatGPT plans might receive it, what email access it could have, and when it might launch, remains unconfirmed.

Editor’s note: This article originally appeared on our sister publication, TechRepublic.