According to media reports on September 19th， cybersecurity firm Radware’s researchers have disclosed that OpenAI has successfully patched a security vulnerability in ChatGPT that could have been exploited by hackers to steal user Gmail email data. This incident highlights a critical concern for the security of integrated third-party services within AI platforms.
The vulnerability was identified within the “Deep Research” proxy tool， which was launched in February of this year. This tool is designed to assist users in analyzing and processing large volumes of information， a feature that， while powerful， presents new avenues for potential exploitation if not rigorously secured.
Researchers indicated that attackers could have potentially leveraged this flaw to pilfer sensitive information from Gmail accounts linked to ChatGPT services. Both enterprise and individual users were potentially affected， with some users possibly facing data exfiltration risks without their knowledge. The implications of such a breach are significant， as it could expose confidential communications， personal details， and even corporate data.
An OpenAI spokesperson responded， stating that model security is a core concern for the company and that they are continuously refining relevant technical standards to enhance the system’s resilience against similar attacks. This proactive stance is crucial for maintaining user trust and ensuring the responsible development and deployment of AI technologies.
