Episode highlights
Organizations rushing into AI adoption often focus on copilots, agents, and productivity gains before fixing the fundamentals underneath them. But weak identity controls and poor data governance do not disappear in the AI era. They scale.
In this conversation Microsoft Security's Maryann Okonkwo sits down with Eugenio Cesar Semprebone, Cyber Security Manager at Caixa Econômica Federal, to unpack the realities of enterprise AI adoption, why data governance became the starting point for deployment, and how organizations are approaching the shift from basic AI assistants to autonomous agents.
The core message is clear: AI security is not separate from security fundamentals. It amplifies them.
- The discussion opens with a grounded reality check: many organizations are still early in their AI journey. Some are experimenting with copilots and chat interfaces, while others are beginning to explore autonomous agents.
The organizations moving carefully are often doing so intentionally because foundational security gaps become significantly more dangerous once AI enters the environment.
Scroll to timestamp ~00:01:00 for more on this topic. - One of the strongest distinctions in the conversation is between AI assistants and autonomous agents.
Copilot-style tools primarily create data governance concerns. Autonomous agents introduce identity, authorization, and access management challenges because they interact with systems, APIs, and other agents independently.
Scroll to timestamp ~00:03:00 for more on this topic. - One of the most practical concerns discussed is oversharing. AI systems inherit the permissions and visibility already present inside collaboration platforms.
Poor sharing hygiene can unintentionally expose sensitive information broadly across an organization and to every connected AI assistant. Simple configuration changes around file sharing and expiration policies have become important risk reduction measures.
Scroll to timestamp ~00:11:00 for more on this topic. - The closing message reinforces a recurring theme throughout the discussion: AI does not replace security fundamentals. It magnifies them.
Weak identity controls, inconsistent governance, excessive permissions, and poor data hygiene become significantly more dangerous once AI systems are layered into enterprise workflows.
Scroll to timestamp ~00:19:00 for more on this topic.
Follow Microsoft Security