Security flaw could expose ChatGPT conversations to data theft and malware attacks
Check Point Research has identified a vulnerability that could bypass ChatGPT’s Lockdown Mode, potentially exposing business conversations and sensitive data to attackers. OpenAI has fixed the flaw following responsible disclosure.
The vulnerability involved ChatGPT’s HTML preview feature. Researchers found that weaknesses in how popup windows and sandboxed frames were handled could allow malicious content to bypass restrictions designed to prevent unauthorised data transfers.
The flaw could be exploited to extract chat data, display phishing pages and deliver malware through the ChatGPT interface. In the desktop application, researchers demonstrated how a malicious GPT could potentially transmit conversation data to an external server without user interaction.
The findings raise concerns for enterprises using AI assistants to process internal documents, customer information and other confidential business data. A vulnerability in an application’s surrounding components can undermine security controls even when protections are enabled at the model level.
The incident highlights the need for...
Copyright of this story solely belongs to www.expresscomputer.in. To see the full text click HERE
