First time Open AI has shut down model due to cybersecurity concerns

“Astra” is OpenAI’s first model to reach what it calls a “critical threshold” in cybersecurity. Evaluations show that the model can hack real systems independently.

It’s the first time that OpenAI has suspended work on a model due to its intrinsic features. Astra has passed a so-called “critical cyber capability” threshold, the first time this has happened. Other research groups will likely encounter the same situation shortly.