Lakera launches to protect large language models from malicious prompts Least News Security Lakera launches to protect large language models from malicious prompts