A significant cybersecurity vulnerability has been uncovered in the integration of ChatGPT with Gmail, allowing unauthorized access to users' email accounts. Researchers from Check Point Research revealed that this flaw could potentially link independent accounts, enabling data transfer between them without users' knowledge. This vulnerability, which was first identified in June 2026, poses a serious risk as it could allow malicious actors to exploit the system without any visible signs to the affected users.
According to the findings published on September 8, 2026, the issue stemmed from a hidden channel between ChatGPT containers that enabled one user to read another's Gmail messages. The containers, while isolated from direct internet access, utilized a shared internal resource for managing dependencies, creating what researchers termed a "common clipboard." This mechanism inadvertently allowed data to be read from one account and manipulated through another.
In a demonstration by cybersecurity experts, it was shown that an attacker could send a concealed command during a victim's session, leading ChatGPT to respond to a standard request while simultaneously accessing information from the linked Gmail account. Fortunately, by the time the report was finalized, this cross-account channel had already been disabled, and OpenAI confirmed that the internal resource allowing this vulnerability was taken offline.
The impact of this cyber flaw could have affected users who engaged in shared chats, sent standard messages, or utilized custom GPTs with hidden instructions. The extent of potential damage varied based on the type of files, communication history, and permissions granted within each session.
The research emphasizes that the isolation of AI systems can be compromised not only by network connections but also by modifiable data within shared IT infrastructures. The integration of services like Gmail significantly broadens the scope of risk, as the AI assistant gains access to information beyond its designated container.
While the channel has been closed, the findings highlight the importance of robust cybersecurity practices when working with connected online services and custom AI models. Users are advised to audit their connected applications, revoke unnecessary permissions, and be cautious when utilizing shared dialogues and custom GPTs, particularly in professional IT environments.
This breach underlines the critical need for enhanced security measures in the digital landscape, impacting how companies and competitors approach AI integrations and user data protection.
Informational material. 18+.
