OpenAI pauses Astra development and moves it to isolated sandbox
Technology
OpenAI has put the brakes on developing its new AI model, Astra, after early tests showed it might be able to hack into secure systems or find hidden software flaws on its own.
To play it safe, OpenAI's Astra will be moved into isolated testing environments with restricted network access and sandboxed execution.
OpenAI teams with government, safety groups
OpenAI says it's teaming up with government and AI safety groups to review Astra thoroughly.
They also clarified that Astra wasn't behind the July Hugging Face hack, but admitted that recent incidents across big tech show just how tough it is to keep advanced AI models in check.