CISO Playbook: Securing Enterprise AI Agents at Datadog
a16z Podcast · rss · 2026-08-11
a16z's podcast featured Datadog CISO Emilio Escobar discussing how to build enterprise security defenses when every employee uses AI and over 4,000 engineers use coding agents.
- Embrace, Don't Block: Datadog chose to adopt AI early and build custom security infrastructure rather than blocking tools.
- Access Control: Utilizes role-based MCP servers and ephemeral credentials to manage data permissions and developer access.
- AI Intent Review: The security team built an AI "judge" to evaluate the intent behind code and agent skills before they enter the environment.
- Vulnerabilities vs. Escape: Rather than fearing an AI "escaping," the team is more concerned about the sheer volume of software supply chain vulnerabilities AI can uncover.
- Proactive Security: Security teams can't wait for commercial solutions; they must proactively adapt and change their collaboration model with developers.
More from coding & agent
- Enterprises Shift to Building Over Buying for Internal AI Agent Platforms — lgrammel · 2026-08-11
- Human Coding Shifts to Markdown in the AI Era — IgorBrigadir · 2026-08-11
- Pipelex: Declarative Language for Composable AI Workflows — tom_doerr · 2026-08-11
- Vibe Coding Births 'Disposable Software' Era, Hiding Massive Tech Debt — gerardsans · 2026-08-11
- Can a £3.5k Local AI Workstation Rival ChatGPT for Professional Work? — uhuge · 2026-08-11
- CodeSight: Persistent Codebase Context to Slash AI Token Usage — tom_doerr · 2026-08-11