Confidential-AI Readiness Checker
Scans a sample deployment configuration against the confidential-computing production checklist and returns a prioritized remediation plan, extending the confidential-computing-inference field note.
Deterministic scan + AI narrative — sample configs, not a live scanner
The pass/partial/fail result for each checklist item is looked up deterministically from a fixed record for the sample you select, not judged by an AI model — a security readiness check should not be able to hallucinate a pass. Only the executive-summary narrative and remediation order underneath is AI-generated, and it is instructed to narrate the given results, not re-judge them. The three sample deployments are synthetic and illustrative, not a scan of a real system.
How it works
The six checklist items mirror the production checklist from the confidential computing field note ↗: confidential VM SKU, remote attestation before secret release, memory encryption in use, network isolation, no plaintext secrets in the guest image, and attestation-bound key release. Pick one of three sample deployments to see how it scores.
Try it
Select a sample deployment to scan against the checklist.
Loading samples…
Pick a sample above to begin.
Why this exists
This product extends the Cloud Capacity project record's theme of rigorous, evidence-led infrastructure decisions: before treating a deployment as confidential-computing-ready, this walks the same checklist a real review would, in the same order of priority — attestation and key-release binding before network hardening, since without attestation nothing else can be trusted.
Production checklist
- Checklist scoring is a deterministic lookup, not an LLM judgment — no hallucinated passes.
- The AI layer only narrates already-computed scan results; it cannot alter any item's state.
- Remediation ordering follows real dependency logic (attestation gates trust in other items).
- Dedicated Foundry prompt agent, versioned and isolated from the portfolio chatbot.
- Automated backend tests cover the fully-ready and mostly-failing sample cases.