Cloudflare AI Vendor Risk Profile
Edge network and security company that operates Workers AI, an inference platform running open-weight models on its global network. Hosts models from Meta, Mistral, and others directly on edge nodes, while providing AI Gateway for routing and observability across any AI provider.
Risk overview
Risk score: 26/100
Risk tier: Moderate
Safety rating: 74/100 (higher is better)
Lower risk scores indicate lower assessed risk.
Last verified: Apr 19, 2026 Aging
Risk dimensions
| Dimension | Risk score |
|---|---|
| Data Handling | 28/100 |
| IP Exposure | 26/100 |
| Jurisdiction | 13/100 |
| Security | 18/100 |
| Regulatory Compliance | 35/100 |
| Transparency | 55/100 |
| Business Stability | 15/100 |
| Dependency Chain | 26/100 |
| Agent Governance | Not assessed |
Analyst summary
Rating: Recommended
Cloudflare Workers AI provides edge-deployed foundation model inference with the company's mature compliance posture (SOC 2, ISO 27001/27018/27701, FedRAMP Moderate, HIPAA). No-training guarantees, no default logging, and Cloudflare's public-listed transparency reporting make it one of the cleaner AI infrastructure options available.
Bottom line: Strong choice for edge AI inference with clean privacy defaults; a good non-hyperscaler diversification option.
Strengths
- No training on customer inputs or outputs; no default logging of prompts or responses
- Full compliance stack: SOC 2 Type II, ISO 27001/27018/27701, FedRAMP Moderate, HIPAA BAA
- Edge deployment across 330+ global locations for low-latency inference
- Publicly listed (NET) with mature transparency reporting on government data requests
- Multiple foundation models available (Llama, Mistral, DeepSeek, others) under one API
Concerns
- Workers AI model selection leans toward open-weights rather than frontier proprietary models
- Some hosted models (e.g., DeepSeek) carry upstream risk inherited from model origins
- FedRAMP is Moderate (not High); not sufficient for all classified workloads
- Smaller AI ecosystem than hyperscaler peers (AWS, Azure, Google Cloud)
Best for
- Developers already on Cloudflare Workers wanting native edge AI inference
- Workloads needing low-latency global distribution with privacy defaults
- Organizations prioritizing vendor diversification away from hyperscalers
Avoid if
- You need frontier proprietary models (GPT-5, Claude, Gemini) directly
- You need FedRAMP High authorization (not yet available)
- Your workloads require deep MLOps tooling beyond inference
Citations
- Data Handling — Data Retention Period
Cloudflare does not retain prompts or responses from Workers AI except as required to provide the service, and does not log them by default.
- Data Handling — Hipaa Baa Available
Cloudflare offers HIPAA Business Associate Agreements for Enterprise customers covering eligible services.
- Data Handling — Trains On User Data
Cloudflare Workers AI does not use your inputs or outputs to train our models or those of our model partners.
- Jurisdiction Profiles — Incorporation Country
Cloudflare, Inc. is a Delaware corporation headquartered in San Francisco, California (SEC Form 10-K).
- Security Compliance — Fedramp Authorized
Cloudflare for Government holds FedRAMP Moderate authorization, with Workers AI available in authorized regions.
- Security Compliance — Gdpr Compliant
Cloudflare complies with GDPR and offers a Data Processing Addendum with standard contractual clauses for all customers.
- Security Compliance — Iso 27001
Cloudflare holds ISO/IEC 27001, 27018, and 27701 certifications covering its global infrastructure and services.
- Security Compliance — Soc2 Type2
Cloudflare maintains SOC 2 Type II attestation with reports available to customers under NDA.