A large team of computer engineers and security specialists at AI app maker Anthropic has developed a new security system aimed at preventing chatbot jailbreaks. Their paper is published on the arXiv preprint server.
Constitutional classifiers: New security system drastically reduces chatbot jailbreaks
