Hiding instructions for AI is leading to new vulnerabilities, like this one where attackers hide malicious scripts in email that Google Gemini executes.
Expressing Stigma and Inappropriate Responses Prevents LLMs From Safely Replacing Mental Health Providers
