From Europe’s AI Omnibus compromise to Claude’s blackmail fix, self-replicating AI hacks, and vibe coding security failures — this week confirmed that AI’s risks are no longer theoretical.