Hello!

Welcome to Secure Prompt’s weekly newsletter, issue #7.

From Microsoft 365 Copilot’s Mermaid injection exploit to OpenAI’s Tainted Memories flaw in ChatGPT Atlas, this week proves that even the most advanced AI systems remain dangerously exposed. Ransomware groups are now automating encryption and EDR evasion with AI, while deepfake-driven election interference surges globally. Meanwhile, researchers unveil CompressionAttack, showing that even prompt compression can be weaponized - a reminder that in AI security, every optimization hides a potential threat.