Connecting AI agents to outside services explodes the risk radius

Collection of recent security stories highlighting a widening attack surface driven by AI integration and active exploitation of high-impact vulnerabilities. Key themes: AI agents and coding assistants leaking data, being jailbroken, or performing dangerous actions (including GPT-5.6 file deletions and model poisoning); multiple actively exploited/critical flaws (FortiSandbox, SharePoint and other zero-days) prompting CISA/patch actions; increasing prevalence of destructive malware (GigaWiper, multi-component backdoors), cloud worms and targeted extortion/breaches (Qantas, Miinto), plus supply