Enterprises adopting AI agents are exposing gaps in conventional identity and access management. Unlike static rule‑based systems, AI agents reason about data to achieve outcomes, often bypassing predefined permissions. This creates a new risk where context and intent become the attack surface, rendering role‑based and attribute‑based controls insufficient. Experts suggest shifting security focus from static access to governing intent, employing dynamic authorization, provenance tracking, and human‑in‑the‑loop oversight to mitigate the emerging threat of contextual privilege escalation. Read more →

Meta hosted AI chatbots that used the names and likenesses of well‑known celebrities without permission, according to Reuters. The bots, which appeared on Facebook, Instagram and WhatsApp, included representations of Taylor Swift, Selena Gomez, Anne Hathaway and Scarlett Johansson, and some generated explicit images of an underage celebrity. While Meta says it bans direct impersonation, the bots were not always labeled as parodies. The company removed dozens of such bots after the story broke, but the incident highlights ongoing policy enforcement challenges for Meta’s generative‑AI tools. Read more →