docs(blog): governing AI agents on Kubernetes - #188
Merged
Conversation
Evidence-backed post for a cloud-native/security audience. Covers the sandbox, sealed egress, and offline-verifiable audit controls, the runtime-agnostic design, and the open-source anf-mcp token-reduction server. Sourced numbers with dates, honest boundary labeling, mermaid architecture diagram. Draft for review before external publishing. Signed-off-by: Shreyansh Sancheti <43677304+shreyanshjain7174@users.noreply.github.com>
There was a problem hiding this comment.
Pull request overview
Adds a draft blog post aimed at cloud-native security teams, explaining why agent deployments get blocked at review time and how this operator’s controls address that gap.
Changes:
- Introduces a new long-form post framing the “security review gap” with dated industry and incident claims.
- Describes 3 governance controls (gVisor sandboxing, default-deny egress, offline audit verification) plus per-agent cost attribution.
- Adds a mermaid architecture diagram and a section on ANF-based token reduction via
anf-mcp.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+12
to
+16
| - Gartner (Jun 2025) predicts over 40% of agentic AI projects get cancelled by | ||
| end of 2027, and names inadequate risk controls as a driver. | ||
| - An AIRQ assessment via Help Net Security (Jun 2026) found only 11 of 100 | ||
| production agents passed a security bar. 98% carried the "lethal trifecta": | ||
| private data access, exposure to untrusted content, and outbound actions. |
Comment on lines
+50
to
+54
| **2. Network access is declared, not discovered.** The operator ships | ||
| default-deny egress and a Cilium FQDN allow-list, generated per workload. If a | ||
| destination is not in the manifest, the agent cannot reach it. The manifest is a | ||
| reviewable object, so security signs off before anything ships. | ||
|
|
| agent did months later, inside an air-gapped cluster. No "trust me." See | ||
| `cmd/audit-verify`. | ||
|
|
||
| **Plus cost per agent.** When the bill spikes, you know which workload did it. |
Comment on lines
+123
to
+124
| - Operator: github.com/Clawdlinux/agentic-operator-core | ||
| - ANF and the MCP server: github.com/Clawdlinux/agent-native-format |
shreyanshjain7174
added a commit
that referenced
this pull request
Jul 22, 2026
Evidence-backed post for a cloud-native/security audience. Covers the sandbox, sealed egress, and offline-verifiable audit controls, the runtime-agnostic design, and the open-source anf-mcp token-reduction server. Sourced numbers with dates, honest boundary labeling, mermaid architecture diagram. Draft for review before external publishing. Signed-off-by: Shreyansh Sancheti <43677304+shreyanshjain7174@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Draft blog post for a cloud-native and security audience, grounded in the booth materials and the shipped work.
Covers:
Honesty: mirrors the pitch's boundary labeling. No invented numbers, sources dated.
Note: this is a draft for review. It is NOT for direct CNCF self-publish (that channel is editorially gated). Intended homes: our site / dev.to / Hashnode now, plus a KubeCon CFP or CNCF ecosystem guest pitch as a separate track.