Unlocking GPT Security: RHEM Labs’ Revolutionary Approach is Gaslighting AIs

Author(s): Dr Richard Matthews – Chief AI Officer RHEM Labs

Originally published on Towards AI.

Top highlight

Photo by Philipp Katzenberger on Unsplash

As the Chief Artificial Intelligence Officer at RHEM Labs, I’ve spearheaded the development of a prompting methodology which not only enables us to iterate quickly when we have a new idea, but it does so in a secure manner.

GPTs (not generative pre-trained transformers but the OpenAI tools) are vulnerable to many various cyber threats.

None are more important than those aimed at revealing a prompt.

The technique we use is central to enhancing GPT security. It becomes crucial in the face of the inherent vulnerabilities associated with these AI systems.

However, there’s still a significant challenge that we must acknowledge: the effectiveness of this method is compromised when the traffic between the GPTs and their client is transmitted in plain text, especially given the nature of REST APIs through which these interactions commonly occur.

I should know.

It’s how I hacked Snapchat’ SnapMap back in 2018.

REST APIs are fundamental to the operation of web services. This includes those who interact with GPTs. The internet relies on HTTP protocols for communication.

While inherently flexible and widely used, these APIs often transmit data in plain text over the network unless specifically secured.

This presents a notable vulnerability: if this data includes sensitive information… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI