According to @AnthropicAI, Anthropic released next generation Constitutional Classifiers to protect large language models against jailbreaks, applying its interpretability research to make protection more effective and less costly than before, as stated in its research announcement source: https://www.anthropic.com/research/next-generation-constitutional-classifiers and source: https://twitter.com/AnthropicAI/status/2009739650923979066. Key takeaways for traders from the source are stronger jailbreak defense and lower safety overhead explicitly claimed by Anthropic source: https://www.anthropic.com/research/next-generation-constitutional-classifiers and source: https://twitter.com/AnthropicAI/status/2009739650923979066.