According to @DeepLearningAI, Meta announced LlamaFirewall, an open-source toolkit designed to protect LLM agents from jailbreaking, goal hijacking, and exploitation of vulnerabilities in generated code. Source: DeepLearning.AI tweet https://twitter.com/DeepLearningAI/status/1967986588312539272; DeepLearning.AI The Batch summary https://www.deeplearning.ai/the-batch/meta-releases-llamafirewall-an-open-source-defense-against-ai-hijacking/ The toolkit is free to use for projects with up to 700 million monthly active users, as stated in the announcement. Source: DeepLearning.AI tweet https://twitter.com/DeepLearningAI/status/1967986588312539272; DeepLearning.AI The Batch summary https://www.deeplearning.ai/the-batch/meta-releases-llamafirewall-an-open-source-defense-against-ai-hijacking/

According to Anthropic (@AnthropicAI), the company has unveiled new research focusing on 'Constitutional Classifiers' aimed at defending against universal jailbreaks. This research is crucial for trading algorithms relying on AI systems, as it enhances security measures against unauthorized access and manipulation. The paper, accompanied by a demo, challenges users to test the system's robustness, potentially impacting AI-driven trading strategies by ensuring more secure and reliable operations.