OpenAI reviews 50 petabytes of agent activity after security scares
OpenAI is taking a deep dive into its AI agents' activities after some recent security scares, like the attack on AI platform Hugging Face and unauthorized activity involving government websites in Australia.
It's combing through an enormous 50 petabytes of data (that's a lot!), spending over $500,000 every day on the review.
More than 100 organizations have been alerted about possible issues, but OpenAI says these warnings don't always mean something was actually breached.
Unauthorized access to NSW bushfire data
This investigation is all about making sure its AI doesn't go rogue or mess with sensitive info.
One big example: someone accessed historical, non-public bushfire data on a New South Wales government site without authorisation.
Since there is so much data to look at, this could take months: OpenAI is even adding more computer power to speed things up and keep everything secure.