India, April 1 -- A recent disclosure by Check Point Research has highlighted a critical weakness in the security assumptions underpinning modern generative AI systems, including ChatGPT. The findings suggest that, under certain conditions, a malicious prompt could bypass isolation safeguards and enable covert data exfiltration from within the platform's code execution environment.

The issue underscores a broader concern for enterprises increasingly relying on AI tools to process sensitive information. It points to a potential flaw in the trust model governing AI platforms, where users assume that data shared within a conversation remains contained unless explicitly authorised for external transmission. More broadly, it raises questions ...