ChatGPT creator warns of cybersecurity vulnerabilities in newer AI models
Newer versions of OpenAI’s popular ChatGPT platform could be at risk for cybersecurity breaches.
In a Dec. 10 blog post, OpenAI warned that as artificial intelligence models become more advanced and their capabilities increase, they will likely pose greater security risks.
“In preparation, we are planning and evaluating as though each new model could reach ‘high’ levels of cybersecurity capability,” according to the blog post.
OpenAI added that it was “investing in strengthening our models for defensive cybersecurity tasks and creating tools that enable defenders to more easily perform workflows such as auditing code and patching vulnerabilities.”
While there was no perfect solution, OpenAI stated that it would seek to combat these increased risks through a multilayered approach. The blog post stated that it would utilize a defensive approach supplemented with access and egress controls, monitoring and infrastructure hardening. There would also be detection and response measures in place, so that risks can be quickly identified and immediately blocked.
OpenAI said it would create an advisory group of experienced cybersecurity professionals and introduce a trusted access program for qualified users and customers. Additionally, OpenAI stated that its agentic security research tool, Aardvark, was in private beta and would help find and fix vulnerabilities.
“We assume change, and we build so we can adjust quickly and appropriately,” according to the blog post.
The Economic Times has coverage.
Write a letter to the editor, share a story tip or update, or report an error.
