ChatGPT Flaw Allowed Attackers to Silently Extract Gmail Data

Security researchers have uncovered a vulnerability in ChatGPT that could allow an attacker to quietly access information from a user’s connected Gmail account. The issue involved a malicious instruction being planted in a ChatGPT conversation, potentially causing the system to perform actions that were not obvious to the person using it.

According to Check Point Research, the vulnerability demonstrated how an attacker could manipulate ChatGPT into handling sensitive information while continuing to respond to the user normally. The researchers described the problem as a covert data-leakage risk involving connected applications and communication channels within the AI environment.

The discovery highlights a broader challenge facing AI assistants that can interact with external services. When an AI system is given access to email, documents or other personal information, malicious instructions hidden within content can create risks that traditional security protections may not always detect immediately.

Check Point said the vulnerability was reported to OpenAI and that the issue was fixed in February 2026. The incident has nevertheless raised wider questions about how AI systems should isolate user information and control interactions between separate accounts and connected services.

The case underscores the importance of treating AI assistants as part of the modern cybersecurity environment rather than simply as chat applications. As AI tools gain greater access to personal and workplace data, researchers and technology companies are increasingly focusing on stronger isolation, monitoring and safeguards to prevent information from being misused or exposed.