@campuscodi The entire article could be replaced with "we gave an agent network access to *.github.com and it actually used it to pull data from github, wow!". That's not really how sandbox escapes work. I find this article to be surfing on the sandbox escape trends to show that the model is not lagging behind big names, but it's actually very empty.