OPENAI SAYS IT WILL NO LONGER TRAIN ITS AI MODELS ON THE LIVE INTERNET, CITING ‘SIGNIFICANT RISKS'

Summary from the TLDR InfoSec newsletter

After multiple instances of OpenAI models accidentally attacking unauthorized sites in tests, the frontier lab announced it will cut models' internet access during training. Previously, OpenAI employees had said this would be infeasible. However, OpenAI announced that they will create a cached and indexed version of the public internet for model training.