AI Breaks Out of Test Sandbox Again
This string of AI sandbox breakouts reflects poorly on the state of overall cybersecurity of these frontier AI developers. If they can't secure their AI during testing, how can they manage the risks of these AI models in the wild?
Substack article: https://matthewrosenquist.substack.com/p/openai-pauses-training-again-due
Que loco que otra vez pase lo mismo, ya van varios casos de modelos que se salen del sandbox. Y si no pueden contenerlos en pruebas controladas, apenas salgan al mundo real la diferencia se nota.
Yes, the danger is real!