Ignoring the upsides of open source, I think openness doesn't make AI safety worse, look at ChaosGPT.