Secure AI with Cloudflare. Stay in control of your data
Your people are already using AI tools like ChatGPT, Claude and Gemini, often without IT knowing. FirstNet and Cloudflare give you visibility, guardrails and governance, so you can say yes to AI without risking your data, compliance or reputation.
AI is already at work. Is it under control?
Shadow AI
Staff paste customer records, contracts and source code into AI tools IT hasn't approved, and that data leaves your control.
AI agents on your data
AI agents now act on company data and systems. Someone has to decide what each agent can reach, and what it may return.
POPIA and ISO still apply
Using AI does not pause your obligations. Personal information still needs safeguards, and auditors still expect access control and logs.
How we help. Four ways to secure AI on Cloudflare
FirstNet designs, deploys and runs each of these on Cloudflare, alongside our Cloudflare WAF, DDoS and Zero Trust service. When prompts and data must stay in South Africa, FirstNet Sovereign AI runs AI on our own local infrastructure.
Secure AI usage
Cloudflare SASE controls for every AI tool your people use.
- Discover shadow AI: which AI apps are in use, by whom and how often
- Allow, block or redirect access to AI apps by identity
- Inspect prompts in real time to stop sensitive data, source code and regulated information
- DLP on uploads and prompts
- AI security posture management for your approved AI platforms
AI Gateway
One control point for every AI provider your applications call.
- Smart routing with retries and fallback to another provider
- Caching to cut cost and latency on repeat requests
- Rate limiting to protect your apps and your budget
- Full analytics and logs for every request
- A single view of AI spend across providers
Governed AI agents
MCP server portals give AI agents one secured way into your tools.
- A single, secured entry point to internal and third-party tools
- Sign-in through your identity provider: Microsoft Entra ID, Okta or Google
- Curate which tools each team's agents can use
- Full logging of agent activity for audit
Build AI applications at the edge
Cloudflare Workers for AI applications, with security built in.
- Serverless compute and storage on Cloudflare's global network
- Vector search and AI models for retrieval and agents
- WAF, DDoS protection, bot management and API protection in front of every app
AI governance is data governance
AI governance now sits inside your data and security governance, not beside it. The controls on this page produce the access rules, logs and evidence your risk owners and auditors already ask for.
These controls support compliance; they do not certify it. The outcome still depends on your policies, configuration and how the controls are run.
POPIA
Supports the security safeguards POPIA expects for personal information (section 19), and helps you control what is sent to AI providers outside South Africa (section 72).
ISO/IEC 27001:2022
Supports Annex A controls for access control (5.15), cloud services (5.23), secure authentication (8.5), data leakage prevention (8.12), logging (8.15) and network security (8.20).
ISO/IEC 42001
Helps align with the AI management system standard: AI risk assessment, monitoring of AI systems and data governance for AI.
Defence in depth. Governance lives where the data flows
Identity and Zero Trust access
People and agents prove who they are through your identity provider before they reach any AI tool or company system.
Data protection in the application
Sensitive data such as ID numbers and contact details is detected and redacted inside the application or tool, before a response leaves.
Network-layer DLP
Cloudflare Gateway inspects traffic from managed devices and blocks or logs sensitive data sent to AI apps.
Monitoring and audit trail
Every request, policy decision and agent action is logged, giving you an audit trail and evidence for reviews.
How engagement works. From discovery to managed service
Discover
AI usage discovery and risk assessment: which tools, agents and data are in play, and where the gaps are.
Design
Policy and guardrail design: approved tools, data rules and access per team, agreed with your risk owners.
Deploy
Deployment and integration with your identity provider, network, SIEM and applications, logging first and blocking after.
Manage
Ongoing monitoring and managed service: policy changes, reporting and escalation into Cloudflare.
Secure AI with Cloudflare questions
Straight answers from the FirstNet team.
Read every answer in the Knowledge HubWill securing AI with Cloudflare block my staff from using AI?
No, unless you want it to. The aim is to let people use approved AI tools safely, and stop sensitive data rather than whole tools.
Policies can allow, isolate, redirect or block each AI app per user and group, and prompt inspection catches sensitive data on the way out. Most organisations start by seeing which AI tools are in use, then steer people towards approved ones.
Does Cloudflare AI security work with Microsoft 365 and Entra ID?
Yes. Policies use your Microsoft Entra ID users and groups, and people sign in with the work accounts they already have.
Okta and Google identity are supported too. FirstNet connects your identity provider during deployment, so AI access rules follow the same identities as the rest of your security.
Do we need to install software on devices to secure AI use?
Some controls need it and others do not. Inspecting AI app traffic from laptops uses the Cloudflare WARP client on managed devices.
AI Gateway, MCP server portals and data protection built into your own applications need nothing installed on the device. FirstNet recommends the right mix for managed, unmanaged and contractor devices.
How does securing AI with Cloudflare help with POPIA?
It helps you apply POPIA's security safeguards to AI use: see where personal information goes, stop leaks and keep the logs.
You can see which AI tools handle personal information, stop it being pasted into unapproved tools, control what is sent to AI providers outside South Africa, and keep logs that show what happened. These controls support compliance; they do not replace your own POPIA programme, and the outcome depends on how your policies are set and run.
Can we start small with AI security?
Yes. Most organisations start with an AI security assessment and shadow AI discovery for one team, then expand.
From there you add prompt and data controls, AI Gateway for your own applications, or MCP server portals for AI agents as the need grows.
What is shadow AI, and how can a business find it?
Shadow AI is staff using AI tools like ChatGPT, Claude or Gemini without IT approval. Cloudflare shows which AI apps are used and by whom.
Each AI app is listed as approved, unapproved or under review, with usage trends per team. Most organisations run this discovery for one team before they set any blocking policy, so nobody is cut off on day one.
How can we stop staff pasting sensitive data into ChatGPT and other AI tools?
Use prompt inspection and DLP. FirstNet sets up Cloudflare to check prompts and uploads in real time and stop sensitive data reaching AI.
People keep using the approved AI tools; what is stopped is the data that must not leave, such as ID numbers, customer records, financial data and source code. For your own AI apps and agents, redaction can also run inside the application, on any device.
What is Cloudflare AI Gateway, and why would a business use it?
Cloudflare AI Gateway is one control point between your apps and the AI providers they call: retries, caching, rate limits, logs and spend.
If a provider is slow or fails, requests are retried or sent to another provider, repeat questions come from cache, and you get one view of usage and cost across every AI provider your applications use.
Should we secure AI with Cloudflare or use FirstNet Sovereign AI?
Often both. Cloudflare secures the AI tools and agents you already use; FirstNet Sovereign AI runs AI on FirstNet's own South African platform.
Choose Cloudflare controls for visibility, guardrails and logs on public AI tools and your own AI apps. Choose Sovereign AI when prompts, data and inference must stay in South Africa. Cloudflare policies can steer sensitive work to the sovereign platform.
From the Knowledge Hub
In-depth answers about Secure AI with Cloudflare
- What is AI security posture management (AI-SPM)?
- How does a FirstNet Secure AI with Cloudflare project work?
- What is an MCP server portal, and how does it secure AI agents?
- What is ISO/IEC 42001, and does it apply to businesses that use AI?
- Can FirstNet help us build secure AI applications on Cloudflare Workers?
- What is the difference between network DLP and application-layer DLP for AI?
Where security fits
- Sovereign AI AI that stays home, in every layer.
- Voice Every call, clear.
- Security Threats, contained.
- Cloud Data, protected.
- Connectivity Always connected.
Enriched by Sovereign AI: keep AI POPIA-aligned: prompts, documents and data are processed on South African infrastructure instead of foreign APIs. Explore the AI Factory