Why we built Glacis on Cloudflare Workers

Latency matters for operational evidence

Security questionnaires, periodic audits, and policy documents serve a different purpose from event-level operating records. AI workflows may produce many actions between reviews, and each workflow has its own latency, evidence, and legal requirements.

Consequential AI workflows leave little room for evidence overhead. We chose an edge runtime to keep supervision and signing close to the action instead of hauling every event through a central region. Every deployment still measures end-to-end p50, p95, p99, error, and failover behavior against its own latency budget.

Why edge execution matters

Cloudflare’s globally distributed network lets Glacis run compute, storage, and security services close to request paths. That architecture reduces unnecessary distance in the hot path and gives us regional options for deployments with stricter data and resilience requirements.

Keeping the evidence path close to the request can reduce unnecessary cross-region traffic. Shared identifiers and timestamps let operators correlate a signed record with existing transaction telemetry, giving review teams a clearer operational trail.

Security at the infrastructure level

Cloudflare’s managed security services give Glacis a strong infrastructure baseline while we apply product-specific controls above it:

What this means for AI teams

The platform choice gives AI teams a practical set of deployment options:

• Measured overhead: receipt and verification latency tested against the workflow’s budget
• Distributed routing: provider routing and failover capabilities evaluated for the selected services
• Regional control options: architectures that can be configured around data-sovereignty requirements
• Managed security options: DDoS, WAF, and bot controls where selected and configured
• Signed application records: key-based integrity and attribution checks for covered fields; no hardware-attestation claim unless separately implemented and verified

Why the platform choice matters

Cloudflare’s support for distributed traffic, edge execution, and managed security controls gives Glacis a strong substrate for supervision that has to happen in the path of a live AI action.

When selected and configured, those platform properties can support:

• Capacity and failure-mode testing against an agreed workload
• Current provider support, service lifecycle, and security documentation evaluated during diligence
• A distributed execution option rather than a mandatory centralized evidence hop

Primary sources

• Cloudflare Workers: network overview
• Cloudflare Docs: WAF overview
• HHS OCR: HIPAA Security Rule audit controls newsletter

Infrastructure for operational evidence

AI rules and buyer requirements vary by system, role, jurisdiction, and date. Reviewers may ask for operating evidence alongside policies, testing, and assessments; neither Cloudflare infrastructure nor a signed record determines compliance.

Periodic audits and documentation remain useful, but they may not answer what a configured control reported for a particular consequential event. Signed records can add integrity and attribution for covered fields while trusted collection, testing, and coverage evidence address actual execution and effectiveness.

We chose Cloudflare Workers in 2025 to explore a distributed, measurable path for generating and verifying signed operational records. Current architecture, latency, security controls, and data flows must be established from the deployed configuration rather than inferred from this launch post.