Securing the platform: Hard-won lessons from the teams who've been breached, scaled, and survived
summary
Panel argues that platform teams must treat software agents like first-class identities, with golden path provisioning, centralized environment images, and per-agent quotas to avoid runaway machine-speed change. Practical controls: VM boundaries instead of relying on Kubernetes namespaces, an LLM proxy for logging and audit trails, automated CVE checks and fast rollout pipelines, and a five-field intake (human, tool, data, use case, exec sign-off) to triage queues that now reach tens to hundreds of daily reviews.
AI-written summary · about
chapters (13)
- 0:00Security panel introduction
- 1:54Platform as attack target
- 4:00Shadow agents security risks
- 7:11AI governance challenges
- 9:00Kubernetes namespace limits
- 10:49Agent security struggles
- 12:56Secure AI supply chains
- 15:26High-frequency change risks
- 17:54Speed versus security balance
- 20:19AI security mistakes
- 22:26AI agent development lessons
- 23:28Managing AI agents overhead
- 26:15Platform security action plan
related talks
26mPlatform Engineering Loves Security: Shift Down To Your Platform, Not Left To Your Developers! - Maxime Coquerel, Royal Bank of Canada
15mPlatform Engineering Security
41mPanel Discussion: Foundations of Security and Leadership at Scale
27mSecurity: The Thing That Everyone...
35mCloud Native Security for the Rest of Us
14mBuild Cyber-Awareness in Platform Development Teams
21mIncorporating Security into Platform Engineering: The Dos and Dont’s
06mThe End of Localhost: A Platform Team's Guide to Securing Agenti...