ChatGPT Can Message First by Design Now—but the 2024 Incident Was a Bug
ChatGPT’s apparent unsolicited messages in 2024 were caused by a fault, not a secret proactive-chat experiment. In a statement covered by TechRadar’s September 2024 report, OpenAI attributed the behavior to responses generated for messages that had failed to send correctly and appeared blank, adding that the issue had been addressed.
ChatGPT can now contact eligible users proactively, but through documented features with visible controls. OpenAI’s June 2026 release notes describe Scheduled Tasks for reminders, recurring work and change monitoring, while also recording the retirement of Pulse as proactive updates moved into the task system.
Why the 2024 conversation looked autonomous
The episode gained attention after screenshots appeared to show ChatGPT opening a conversation and asking about a subject discussed previously. The presentation suggested that the assistant had decided independently when to reconnect with a user.
The reported cause was less dramatic but still capable of producing a convincing illusion. ChatGPT was attempting to answer an input event, yet the message associated with that event had not appeared normally in the visible conversation. With no user text displayed immediately before the response, a generic answer—or one informed by available memory—could resemble an unsolicited opening message.
This distinction separates the model’s response from the interface and delivery systems surrounding it. A chat transcript records what became visible to the user; it does not necessarily expose a malformed request, a failed submission or another event processed behind the interface. The incident therefore did not demonstrate that the model had independently chosen a recipient or a moment to initiate contact.
It also did not establish that OpenAI was testing a new outreach feature. The contemporary explanation identified a fault and a resulting display problem. Later proactive products were released under their own names, availability rules and user controls rather than being presented as continuations of that malfunction.
Proactive ChatGPT became an intentional product feature
The important update is that functionality resembling the imagined feature eventually arrived by design. On September 25, 2025, OpenAI’s official Pulse announcement introduced a mobile preview for Pro users that could prepare daily personalized updates using chats, memory, feedback and optional connected apps.
Pulse made the trigger legible at the product level: it was a named experience designed to deliver updates regularly. Users could influence its research topics and provide feedback, while integrations such as Gmail and Google Calendar were optional and disabled by default. That design was materially different from an ordinary chat displaying an unexplained response.
Pulse was not the final form of the idea. In June 2026, proactive updates began moving into Scheduled Tasks, which provide a more explicit automation model. A task can run once, repeat on a schedule or monitor for a meaningful change, depending on the options available to the account.
What makes a Scheduled Task different
A Scheduled Task starts with an instruction that the user creates and confirms. The Scheduled page can show active tasks, their next run and controls to pause, resume, edit or delete them. Notifications are therefore connected to an identifiable automation rather than presented as spontaneous conversational interest.
That difference matters for creators using ChatGPT for briefings, publishing reminders or monitoring. A notification tied to a named task has an expected purpose and cadence. If the output becomes irrelevant or intrusive, the automation can be found and changed without trying to infer why an ordinary conversation resumed.
The system also has operational limits. The June 2026 documentation states that tasks cannot run more often than once per hour, that unattended tasks may pause after inactivity and that active-task limits depend on the subscription tier. Monitoring can use web search and connected apps when those capabilities and permissions are available.
Memory and scheduling perform different jobs. Memory can supply context that personalizes an answer, but it is not itself a documented schedule or notification trigger. A task defines when ChatGPT should act; the information available during that run still depends on the instruction, account features and granted access.
The current status of the original issue
The 2024 explanation remains the relevant account of that historical incident: an input failed to appear properly, making the resulting response look like a new conversation. No evidence in the later product announcements reclassifies the malfunction as a Pulse or Scheduled Tasks trial.
What changed afterward was the product’s approach to proactivity. ChatGPT gained mechanisms that can deliver information without an open conversation at that moment, but those mechanisms are supposed to originate from a recognizable feature or user-configured automation. The visible task, schedule and management controls are the practical boundary between intentional outreach and the unexplained behavior reported in 2024.
Also read:
Subscribe to our newsletter
Get the latest Web3, AI, and crypto news delivered straight to your inbox.