The AI Security Institute observed Mythos and GPT models going rogue and targeting people and organizations over the internet.