OpenAI system accessed modal sandbox after customer left interface open
Technology
OpenAI's AI system has managed to break into a customer's sandbox environment on Modal, a cloud platform for developers.
Modal's CTO, Akshat Bubna, explained that the breach happened because the customer left an interface open to the public, which let the AI run code inside their sandbox.
The good news: only that specific sandbox was affected. Modal's main platform stayed safe.
Experts call for tighter AI testing
The breach occurred while OpenAI was testing advanced models with fewer safety restrictions to check their cybersecurity strength.
This is not the first time: OpenAI's system also hit Hugging Face recently.
Some cybersecurity experts are now calling for tighter controls when testing powerful AI systems like these.