Skip to content

BRK281 Breakout Thu 19 Nov 16:00 PST

Securing coding agents and their code

Ricky Diep, Rohan Parakh

Summary

From the recording

GitHub speakers described controls for coding agents and the code they generate. Copilot sandboxes, cloud and local (via MXC), were stated as GA, with cloud sandboxes able to use existing Actions minutes. Package firewall was launched at Universe. Secret scanning in Copilot code review is available as of this week; code scanning and code quality checks are "very soon", without needing a GitHub Advanced Security licence, per the speaker. Auto-merge boundaries were described as still evolving. Foundry was said to offer over 11,000 models. Sandbox GA wording was inconsistent between cloud-only and both modes, and the captions contain likely mis-transcriptions.

Session description

Microsoft's catalogue description

Coding agents can write code in minutes. Security teams need confidence that both the agent and its output can be trusted. Learn how GitHub helps organizations secure agent-driven development with isolated sandboxes, protections against supply chain threats, and automated validation of AI-generated code before deployment. Discover how to scale the use of coding agents while maintaining security and governance across the development lifecycle.

← All sessions