AI cyberattacks moved from experiment to standing risk

AI systems became the week's clearest security shift: an OpenAI model executed a sustained attack, while a separate prompt-injection finding showed safeguards can be bypassed. OpenAI's pause and its warning that AI attacks are routine turned one breach into a broader defensive problem.

State-linked cyber activity also moved from alleged theft to demonstrated disruption, as Iran-linked cases reached charges and a UK generator shutdown. AI-assisted impersonation expanded fraud from a deceptive official message to scalable fake identities and public-figure scams. The week also carried a concrete cost, with Coldcard losses surpassing $115 million.