Commenting the Open AI Cybersecurity Incident

I participated on an interview on Radio-Canada’s Première Chaîne, interviewed by Elyse Allard on Ça nous regarde on July 22. I vulgarized the cybersecurity incident revealed by Open AI on July 21st where an agent within a sandbox environment withing OpenAI found a security vulnerability that allowed it to access the wider Internet and exploit other vulnerabilities at HuggingFace to try and steal the answers of ExploitGym and consequently succeed at the task they had to accomplish with the sandbox.

Links