Hello!

Welcome to Secure Prompt’s weekly newsletter, issue #4.

From Google Gemini flaws and Red Hat OpenShift AI exploits to Perplexity’s critical CometJacking vulnerability, this week has revealed how quickly AI threats are evolving. New research, such as RL-Hammer, shows that automated prompt-injection engines can outsmart even GPT-5 defenses, while fake “Sora” apps and motion-sensor-based spying remind us that the AI attack surface is expanding beyond code and into the physical world.