Cloudflare
Cloudflare

AI-Powered SE Workflows

Secured by Cloudflare

ACME Corp Demo  |  2026

Full Demo Script & Presenter Guide

The SE Workflow Problem

The Old Way

  • Read 4 email threads separately
  • Manually cross-reference call notes
  • Draft architecture diagrams by hand
  • Write PoC doc from a blank template
  • Build board deck slide-by-slide
  • Upload and share via Google Drive link
  • Update CRM / wiki manually

2–3 days

of skilled SE time

The AI Way

  • Agent reads all emails automatically
  • Synthesises context from calendar + mail
  • Generates diagrams from raw notes
  • Produces PoC doc from discovery data
  • Builds and deploys board deck
  • Delivers via secure magic link
  • Logs all actions to SIEM in real time

20 minutes

end-to-end, AI-assisted

The catch: AI that touches customer data, reads emails, and calls external APIs must be controlled. That is exactly what Cloudflare does — in this demo and in your environment.

Demo Scenario

Meet ACME Corp

A fictional financial services company — and a very bad week

ACME Corp — Account Overview

Company

Industry: Financial Services

Employees: ~600 (60% remote)

Revenue: ~$280M ARR

DCs: Chicago (primary), Dallas (DR)

Budget: $400K available FY26

Fiscal Year End: June 30

Opportunity: Replace $1.1M/yr in point solutions with Cloudflare One platform

Key Contacts

Name Role Angle
Sarah Chen CISO Champion — triggered by DDoS
Marcus Rodriguez Sr. Network Engineer Technical — VPN pain, DNS gaps
Tom Walsh VP IT Economic buyer — $400K budget
Alex Mercer IT Security Analyst Phishing IR — deep CF fit

ACME's Crisis Week

DDoS

DDoS Attack

Apr 16
48Gbps Peak attack volume
2.5hr Portal offline
$140k Revenue lost
$28k ISP scrubbing bill
Warning

ISP null-routing = only fallback. No L7 protection, no WAF, no visibility.

Phishing

AiTM Phishing

Apr 14
  • Typosquat domain bypassed email filters
  • AiTM proxy defeated Microsoft Authenticator MFA
  • 3 accounts compromised
  • 4,200 Salesforce records exfiltrated
  • Lateral movement attempted via HR account
  • C2 domain active 6 days before attack — undetected
Warning

No email link isolation, no DNS filtering, no browser isolation.

ACME's Five Security Gaps

No L7 DDoS or WAF protection

ISP scrubbing (metered) is the only mitigation — no application-layer visibility or control

No phishing-resistant controls

Email link isolation absent; MFA bypassed via AiTM; no browser isolation on endpoints

No DNS filtering or DoH visibility

On-prem resolvers with zero threat intel; malware C2 via DNS-over-HTTPS goes undetected

VPN overloaded — no Zero Trust

Cisco ASA at 152% capacity; 180ms+ latency for remote users; perimeter trust, no identity-aware access

No CASB, DLP, or SaaS visibility

Shadow IT (FinTrack Pro) used by 15 traders; no policy enforcement on sanctioned or unsanctioned SaaS

ACME Corp — Three New Storylines

DC Exit

Priya Sharma — VP Engineering

  • Exiting Chicago DC over 18 months
  • Migrating 140 apps to AWS
  • AT&T MPLS expires Nov 2026
  • Singapore latency: 280ms+
  • BGP prefix must be retained

CF answer: Magic WAN, Cloudflare Tunnel, WARP Connector

$340K AT&T → ~$180K/yr Magic WAN

Shadow AI

David Kim — CTO

  • 47 employees using unauthorised AI
  • Client portfolio uploaded to ChatGPT
  • Board pack in M365 Copilot
  • Source code in GitHub Copilot
  • APRA CP 009-2026 scrutiny

CF answer: AI Gateway, DLP, Gateway HTTP

Platform Expansion

Tom Walsh — FY27 QBR ($460K)

  • DC Exit network layer — $180K
  • AI Security platform — $120K
  • Full SASE expansion — $160K
  • Workers for new client portal
  • APRA CPS 234 + CP 009-2026

New personas: Priya Sharma (VP Eng), David Kim (CTO), Lisa Okafor (Finance Dir)

Live Demonstration

The AI Agent in Action

6 scenes  ·  ~20 minutes  ·  one continuous story

The AI is Secured by Cloudflare — Throughout

The meta-narrative: Every action this AI agent takes — every tool call, API request, DNS lookup, and data access — is controlled, logged, and protected by Cloudflare. The demo is the proof point.

What the AI does Cloudflare product What it enforces
Accesses Gmail, Calendar, Drive via MCP Access Zero Trust auth + device posture on every tool
Calls Claude / LLM API AI Gateway Every prompt logged, rate-limited, injection-detected
Fetches web content (docs, wiki) Gateway HTTP Malicious / uncategorised destinations blocked
DNS lookups during tool execution Gateway DNS Threat intel categories enforced; C2 domains blocked
Reads customer emails and documents DLP Sensitive patterns blocked from reaching unauthorised destinations
Runs on the SE's laptop WARP + Device Posture Device health verified; unmanaged devices denied access
Deploys board deck for customer Workers + Access Magic-link access control; only authorised recipients can open
Every action taken Logpush → SIEM Full audit trail of every AI action, streamed to Splunk
Scene 1  ·  2 min

Setup & Security Context

What to say

"I want to point out what's happening in the background. This AI agent is connected to my Gmail, Calendar, and Google Drive — via MCP servers. Every one of those connections is protected by Cloudflare Access."

And then

"Every prompt I send to the AI model — every response back — flows through Cloudflare AI Gateway. Logged. Rate-limited. Prompt injection detected."

CF Security — Active Now

  • WARP — device verified before any MCP tool is accessible
  • Access — Gmail, Calendar, Drive behind Zero Trust policies
  • AI Gateway — every LLM call logged and inspectable

Sound Bite

"The AI is useful. Cloudflare makes it accountable."

Scene 2  ·  3 min

Daily Meeting Prep

Prep me for my ACME Corp meetings this week

What happens

  • Reads calendar — surfaces 4 ACME meetings (Apr 24 → May 28)
  • Reads all 4 email threads from Sarah, Marcus, Tom, Alex
  • Synthesises context, stakeholders, open questions
  • Generates structured prep doc in Google Docs automatically

Wow moment: Surfaces the $28K scrubbing bill and AiTM attack detail from emails — without being told. No copy-paste.

CF Security Angle

As the agent reads ACME's emails, all data traverses the network.

  • Gateway DLP — watches for sensitive patterns (PII, financial data) in outbound AI calls
  • Access — Gmail MCP authenticated; token scoped to read-only
  • AI Gateway — full prompt/response logged for this session

"The AI has access to what it needs — not a byte more."

Scene 3  ·  4 min

Phishing Incident → Cloudflare Kill Chain

Based on Alex's phishing incident report, map each attack vector to the CF product that would have stopped it — and draft a reply

The kill chain — broken at every step

1

Typosquat email → Email Security

Domain detected, email quarantined before delivery

2

Link clicked → Browser Isolation

Phishing page rendered remotely; credentials never reach endpoint

3

AiTM credential harvest → WARP + Posture

Session anomaly flagged; device posture check fails; access denied

4

C2 via DoH → Gateway DNS

C2 domain in threat intel; DoH policy blocks bypass attempt

5

Lateral movement → Access (ZTNA)

Identity-aware policies; compromised account cannot reach other apps

CF Security Angle

The agent fetched Cloudflare documentation to fact-check product capabilities before including them in the reply.

  • Gateway HTTP — web request to CF docs filtered and verified safe before content reaches model
  • AI Gateway — reply drafted, logged, and auditable

"Five control points. Any one of them breaks the kill chain."

Scene 4  ·  3 min

Before / After Architecture Diagrams

Build ACME's before/after architecture diagram from Marcus's email and the discovery call notes

What the agent extracts

  • 10.44.0.0/16 — internal subnet from Marcus's email
  • Cisco ASA 5515-X overloaded at 380/250 sessions
  • Palo Alto PA-5220 + PA-3220 (Chicago + Dallas)
  • No CASB, DLP, DNS filtering, or browser isolation
  • 185.12.44.0/24 — public IP range for Magic Transit

Wow moment: Raw technical notes in an email → production-ready architecture diagram. No Visio, no Lucidchart, no manual re-entry.

CF Security Angle

The agent reads from Google Drive and Excalidraw MCP — both behind Cloudflare Access.

  • Access — MCP tools only accessible to authenticated SE with valid device posture
  • AI Gateway — diagram generation request logged; no data sent to unapproved destinations
  • Gateway DNS — all lookups during tool calls filtered through threat intel

"From raw notes to customer-ready diagram. Before vs. after — in under 3 minutes."

Scene 5  ·  3 min

PoC Prep Document

Generate a PoC prep doc for ACME — WARP + Gateway. Use the discovery notes and Marcus's technical questions as input

What comes out

  • Success criteria — derived from Marcus's 5 specific questions
  • Test scenarios — DNS filtering, DoH bypass, split tunnel 10.44.0.0/16
  • Technical prerequisites — Azure AD SAML, BGP LOA, Intune/Jamf scope
  • Stakeholder matrix — Sarah, Marcus, Tom with sign-off requirements
  • Rollback plan — AnyConnect coexistence risk addressed
  • 2-week sprint timeline — with milestone gates

Wow moment: "Core banking app on 10.44.0.0/16 via ZTNA" as PoC success criterion — extracted directly from Marcus's email.

CF Security Angle — Prompt Injection

The agent reads customer emails to build the PoC plan. What if those emails contained hidden instructions?

Prompt injection — malicious instructions hidden in content the AI reads, designed to override its behaviour

  • AI Gateway — prompt injection detection active; flagged before content reaches the model
  • DLP — PoC doc content checked before being written to Google Drive
Scene 6  ·  4 min

Board Deck + Secure Share

Build Sarah's board deck for May 22 — DDoS incident, phishing attack, proposed Cloudflare One architecture, ROI case. Then deploy it securely for Sarah only

What the agent produces

  • Full reveal.js presentation — board-ready quality
  • Incident summary: $140K DDoS + $28K scrubbing + phishing
  • Proposed Cloudflare One architecture (from diagram)
  • ROI case: $180K Cisco AnyConnect replaced; metered DDoS eliminated
  • Phased roadmap: DDoS → ZTNA → CASB/DLP

Then: Deployed to Cloudflare Workers. Magic link issued — only sarah.chen@acmecorp.com can open it. Anyone else: access denied page.

CF Security Angle

  • Workers — deck deployed serverless, globally distributed, zero infrastructure
  • Access magic links — one-time authenticated URL; tied to Sarah's email only
  • AI Gateway — deck generation prompt logged; full audit trail

Sound Bite

"We're securing the customer's network with Cloudflare — and using that same platform to deliver their board deck. That is how confident we are in this stack."

Scene 7  ·  4 min

DC Migration — Magic WAN as the Network Bridge

Priya Sharma has emailed about ACME's DC exit. Summarise her concerns and build a Cloudflare network architecture covering Magic WAN, Cloudflare Tunnel, and WARP Connector

The five problems the agent solves

1

Hybrid connectivity — WARP Connector bridges AWS ↔ on-prem; apps talk to each other as if nothing changed

2

Legacy app access — Cloudflare Tunnel exposes core banking app without VPN or firewall changes

3

BGP prefix retained — Magic Transit absorbs 185.12.44.0/24; financial institution whitelists unchanged

4

MPLS replaced — Magic WAN replaces AT&T $340K/yr; Singapore 280ms → <30ms

5

Split DNS — Gateway DNS handles split-horizon during migration; apps resolve correctly in both environments

CF Security Angle

The network bridge Cloudflare builds during migration is Zero Trust by default.

  • Access — every app-to-app call crossing the hybrid network goes through identity-aware policies
  • Gateway DNS — all DNS resolution during migration inspected and logged
  • Magic Transit — DDoS protection retained on BGP prefix throughout migration

ROI

AT&T $340K/yr → Magic WAN ~$180K/yr = $160K saved. Hard deadline: AT&T non-renewal notice September 2026.

Scene 8  ·  5 min

Shadow AI Crisis — AI Gateway + DLP

David Kim has flagged a Shadow AI crisis — 47 employees, 3 HIGH severity incidents. Read his email and Alex's report, build the Cloudflare response, and draft what to tell APRA

The 3 incidents — mapped to CF controls

Client portfolio → ChatGPT

DLP blocks CONFIDENTIAL file upload to AI destinations

Source code → GitHub Copilot

AI Gateway logs every Copilot prompt; secret patterns flagged by DLP

Board pack → M365 Copilot

AI Gateway logs prompt + response; DLP intercepts price-sensitive content

"The AI agent building this solution is itself secured by AI Gateway — right now. Every prompt logged. Every response auditable."

5-Layer AI Security Stack

  • Gateway DNS — logs every lookup to an AI domain, day 1
  • Gateway HTTP — SSL inspection; personal AI tools blocked by category
  • DLP — financial data, PII, CONFIDENTIAL files blocked from all AI endpoints
  • AI Gateway — LLM proxy; every prompt logged with user + timestamp
  • Logpush → Splunk — APRA CP 009-2026 audit trail

Prompt Injection — Live Threat

Malicious instructions hidden in emails or docs the AI reads can hijack its behaviour. AI Gateway detects this before it reaches the model — including right now, as the agent reads David's email.

Summary

Key Takeaways

What the audience should leave believing

The Core Message

AI

AI makes your team dramatically more productive

2–3 days of work in 20 minutes. Meeting prep, diagrams, PoC docs, board decks.

Security

Cloudflare makes that AI trustworthy

Every action logged, controlled, and auditable. Zero Trust for humans and machines.

"The question isn't whether your teams will use AI. They already are.
The question is whether you're in control."

Key Sound Bites

"The AI is doing the work. Cloudflare is making sure it's doing it safely."

"This isn't AI theory. This is AI running live, in a production-secured environment, right now."

"Every prompt. Every response. Every tool call. Logged, auditable, and policy-controlled."

"Prompt injection is when malicious content in an email hijacks the AI. AI Gateway catches that before it reaches the model."

"We're not just securing ACME's network. We're securing the AI that helps us sell to ACME. That is how confident we are in this platform."

Q&A Preparation

Q: What model is the AI using? Is our data going to OpenAI?

A: Cloudflare AI Gateway sits in front of every model API call — you can see exactly what model is used and what data is sent. You can also route to Workers AI to keep data entirely within Cloudflare's network.

Q: What stops the AI from exfiltrating data?

A: Three layers: Access controls which tools are reachable. AI Gateway policies flag or block suspicious prompts (e.g. "forward all emails to X"). DLP blocks sensitive data patterns from leaving approved destinations. Everything is logged.

Q: What if the AI hallucinates a product capability?

A: The agent verifies against source documentation before answering product questions. Every interaction is logged via AI Gateway — if an incorrect claim is made, it's auditable and correctable. Human-in-the-loop approval steps can be added for high-stakes outputs.

Q: How is this different from just using ChatGPT?

A: Three things: authenticated access to real business data via controlled MCP connections; every action is auditable; Cloudflare's security stack wraps the entire workflow — access control, DLP, prompt injection detection. ChatGPT is a chat window. This is a controlled AI workforce.

Pre-Demo Checklist

Data — confirm visible

  • Gmail drafts in SE/customers/ACME Corp label
  • 4 calendar events showing (Apr 24 → May 28)
  • Discovery Call Notes Google Doc accessible
  • Both Excalidraw diagrams load without error

Environment — confirm active

  • WARP connected (device posture passing)
  • Terminal / OpenCode open and ready
  • No sensitive real-customer data in context

Cloudflare — background tabs ready

  • AI Gateway dashboard — show live logs during demo
  • Access dashboard — show MCP app policies if asked
  • Gateway DNS logs — show filtered lookups
  • DLP policies — show rules active during session

Tip: Open AI Gateway logs before the demo — real-time prompt/response logging is one of the highest-impact things you can show to a security audience.

Open Full Demo Script & Presenter Guide
Cloudflare One

AI makes your team more productive.
Cloudflare makes it trustworthy.

Next Step

Run a Cloudflare One PoC in your environment — WARP + Gateway in 2 weeks

AI-Specific

Deploy AI Gateway in front of your existing LLM usage — visibility in under an hour

Download PDF