After the first-ever summer break, the crew is back! New crew. New format. Listen or watch to hear about what’s coming. We also welcome Alistair Pugin. Microsoft MVP for M365 + Security, Blogger, Podcaster and Speaker. Key Highlights * Return of the Show (1:38): The hosts are back after a three-month summer break, during which they experienced new jobs, roles, and duties. They thank their listeners and confirm the original cast of Edward Walton, Rod Trent, and Franklin Grimberg are back, though Brody is still on hiatus. * Focus on AI and Security (0:52, 1:02): Frank highlights the current “crazy” world of AI, particularly Microsoft’s efforts to secure and manage it. He expresses concern that many people are unaware of the tools available to them. * Guest Introduction - Alistair Pugan (5:57): Alistair Pugan, from Cape Town, South Africa, is introduced as an expert in compliance and information protection, having worked with Microsoft on shaping exams like SC400 and even co-designing a board game about deception. * Challenges with AI Adoption (7:58): Alistair discusses the “wild wild west” of AI adoption, where organizations are indiscriminately handing out AI, and users are not following guidelines. He notes the parallel to the Google search appliance debacle of 2008, where people are finding content they shouldn’t. * Microsoft’s AI Strategy and Data Training (20:08): The discussion touches on Microsoft’s stance that they do not train their AI models on customer data, emphasizing the importance of data classification for protection. * Copilot as Superized Search (24:15): Alistair explains that Copilot functions as a “superized search” within the Microsoft 365 tenant, using semantic indexing and security trimming to ensure users only access data they have permissions for. * Data Security Posture Management (DSPM) for AI (28:45): The hosts delve into DSPM for AI, a tool within Microsoft Purview (E3 or E5 licenses) that helps organizations monitor their AI usage. Key aspects include: * Components of Data Security (29:51): Frank and Alistair discuss how Microsoft defines data security, including information protection (sensitivity labels), data loss prevention (DLP), and insider risk management. * Monitoring AI Usage (31:25): DSPM allows organizations to monitor what users are doing with AI, including AI usage reports and integration with Defender for Cloud Apps. * Prompt Monitoring (32:28): It can monitor user prompts, especially for sensitive information requests (e.g., “give me the payroll for everyone”), using sensitive information types or trainable classifiers. * Shadow AI Detection (33:21): DSPM helps detect “shadow AI” by monitoring when users visit or upload sensitive information to third-party AI sites like Chat GPT, Gemini, or Perplexity. * Policy Automation (34:31): The tool can automatically spin up policies to detect sensitive information in AI prompts, visits to AI sites, and sensitive data uploads to AI sites. * Agent Sprawl and Non-Human Identities (15:50, 17:10): A significant concern raised is that anyone with a Microsoft 365 Copilot license can build an agent in Copilot Studio, which registers an application in Entra (Azure Active Directory) and creates “non-human identities.” This can lead to “agent sprawl” and uncontrolled API permissions if not properly managed by identity admins. * Mitigating Agent Sprawl (40:03): The solution involves having an application security posture management strategy and robust application onboarding and offboarding policies, as agents are essentially applications that require permissions to interact with data. * Copilot Studio Licensing (39:02): There are different licensing models for Copilot Studio: a free tenant license for building agents (for users without an M365 Copilot license) and a premium capacity license for deploying agents to users without a Copilot license. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.microsoftsecurityinsights.com
