Aug 11, 2026 · 23 min
Datadog CISO shifts AI security from panic to practice
The CISO Playbook for AI Agents | Datadog
Datadog’s experience shows how security teams can govern rapid AI adoption without treating every theoretical failure as an emergency.
- 1Datadog expanded AI access while tightening controls around data permissions, credentials, coding agents, and software supply chains.
- 2Role-based MCP servers and contextual analysis help separate meaningful vulnerabilities from the noise that frustrates developers.
- 3Emilio Escobar argues that security teams should prioritize intent, vulnerability volume, unequal capabilities, and practical collaboration over AI escape scenarios.
Don't miss
Emilio Escobar explains why he is more concerned about vulnerability overload, unequal access to frontier capabilities, and unclear regulation than hypothetical AI escape scenarios.
The brief
Datadog adopted AI broadly, moving from an initial group of Cursor licenses to coding agents and general-purpose tools rather than risk falling behind.
Emilio Escobar says natural-language interfaces exposed weaknesses in warehouse permissions, prompting Datadog to use role-based MCP servers and stronger governance.
Coding agents create a concentrated attack surface: their credentials can reach production, enable package worms, or amplify harmful marketplace skills, so Datadog evaluates code and skills with automated safeguards.
The conversation’s sharpest turn is Emilio’s refusal to panic about AI escape scenarios; malicious actors already exist, while vulnerability volume and unclear rules pose more immediate operational problems.
His practical prescription is to judge intent and context, reduce irrelevant scanner noise, and bring developers, security teams, and developer-experience groups into the same workflow.
Featuring
Datadog
ChatGPT
Gemini in Chrome
a16zListen to the full episode and explore every guest, topic, and moment on PodLume.