Application Security — Noble Demos

DDoS Protection

When the flood comes, you're already behind the dam.

DDoS attacks don't ask permission. They show up in volume, at any layer, at any hour. Cloudflare's network absorbs them automatically — no rules to write, no pages to wake up to. Your origin keeps serving real users while the attack hits a wall it can't break.

Open dashboard ↗

Volumetric flood · absorbed at the edge illustrative · the wall doesn't move

attack volume hitting edge 5.6 Tbps

reaching your origin 0 bps

No attack traffic tax Blocked DDoS traffic doesn't create mitigation overage charges — unmetered DDoS protection is included at every plan tier.

58x headroom

329 Tbps network capacity vs. 5.6 Tbps largest attack on record — autonomous mitigation in seconds, zero service degradation

Predictable billing

DDoS mitigation is unmetered — malicious traffic absorbed at the edge doesn't create overage charges. Pay for clean traffic, not the flood.

Full-stack defense

Network (L3), transport (L4), and application (L7) layer protection — comprehensive coverage at every attack vector

DDoS docs

Security Rules

Defense in depth — managed by Cloudflare, customized by you.

Open dashboard ↗

Layered defense · request → managed → custom → rate-limit → origin → data-scan

Actions per rule: block managed challenge log skip (custom rules)

Managed Rules · Signature-based

We write them. We update them. You stay protected.

Deterministic patterns — regex, string & structural matches — authored by Cloudflare and OWASP.

New signatures ship within hours of a CVE — the network is patched while teams are still triaging.

⚡ Zero-day CVEs (e.g. Log4Shell) Network-wide patch in minutes · industry avg: 60+ days to patch

Custom Rules · Your policy + ML signals

Your rules. ML signals included.

Combine standard request fields — geo, path, headers — with ML-inferred signals like the WAF Attack Score.

cf.waf.score 1 = attack-shaped · 99 = clean

1–20 Attack
21–50 Likely attack
51–80 Likely clean
81–99 Clean

The score informs — you set the threshold and choose the action.

Rate Limiting Rules · Advanced

Count what matters. Stop abuse at the edge.

Count by anything: cookie, API key, JA3/JA4 fingerprint, ASN, JSON fields. Each key gets its own budget.

Count only what matters — 4xx-only for credential stuffing, or complexity-based so a GraphQL bomb costs more than a static asset.

🚪 5 logins/min per IP — block 10m Credential-stuffing fraud: $6B+/yr in e-commerce losses (Forrester)

Sensitive Data Detection

Stop the leak before it lands.

Scans response bodies (text, HTML, JSON, XML up to 1 MB) for PII, financial data, and secrets leaving your origin — including responses from cache and Workers.

Detection runs off the response path — zero added latency. Action is always log (the response has already been sent), so you get full visibility into what slipped through your code.

💳 Credit card numbers (PCI) PCI-DSS non-compliance: $5K–$100K/month in fines per merchant (Visa)

Zero engineering lift

Edge-enforced security with no code changes, SDK integration, or application server dependencies

Global in <60s

CVE patches deployed network-wide in under a minute — closing the exposure window before attackers exploit it

Unified control plane

Managed, custom, and rate-limiting rules configured from a single dashboard — no vendor sprawl

See it live → Security Rules docs

Bot Management & AI Crawl Control

Automation isn't one bucket anymore.

Malicious bots get a score. Sessions get verified. AI crawlers get classified by intent — search, agent, or training. Business value gets surfaced. Then you decide what happens: allow, challenge, or block — all at the edge, no code changes needed.

Open dashboard ↗

The Bot Score illustrative · every request lands somewhere on 1–99

Bot Score

Every request scored 1–99.

ML + heuristics + behavioral signals Score available as cf.bot_management.score — route into WAF rules, rate limits, or transform rules

Threshold rules, not binary blocks Block=1, challenge 2–29, trust ≥30. Tunable per route, no redeploy

Precursor + JS Signals

Catch what static signals miss.

Precursor: session-based detection Flags early-session abuse patterns before bots reach sensitive endpoints

JS detections + browser fingerprinting Catches headless browsers and proxy farms that pass every static check

Precursor trace tool

BotBase + Verified Bots

Allow the good ones. Block the fakes.

15,000+ bot + AI agent signatures Verified by behavior and detection IDs — not just User-Agent headers

Impersonator detection Reverse DNS classification catches bots claiming to be Googlebot or GPTBot

AI Crawl Control

Classify AI crawlers by intent. Act on each.

Search · Agent · Training Dedicated dashboard — allow, block, or manage each intent category independently

Five focused dashboard areas Overview · Metrics · Security · Optimization · Signals

AI Crawl Control Five dashboard areas — one place to understand and act on every AI crawler hitting your zone

01 Overview Operator snapshot

02 Metrics Volume & trends

03 Security Allow · block · manage

04 Optimization Reduce crawl waste

05 Signals robots.txt · policy

Classify by intent: Search Agent Training

Score abuse

ML + heuristics + behavioral signals on every request — before it reaches your origin

Classify intent

AI crawlers sorted by purpose — search, agent, or training — with policy signal context

Act on value

Allow, challenge, or block by category, score threshold, or business value. No code changes.

AI Crawl Control ↗ Bot Management docs

AI Application Services

Protect what AI takes. Protect what users send.

Two traffic directions face different risks. Cloudflare gives each direction its own signals, policy, and outcome.

01 · Govern content access

Know which AI crawler arrived—and what job it came to do.

AI Crawl Control turns a generic bot request into a deliberate content decision.

Open AI Crawl Security ↗

OBSERVE

See demand and value

Compare crawlers, operators, paths, bytes transferred, and referral traffic.

WHO IS TAKING?

DECIDE

Separate Search, Agent, and Training

In Security Settings, allow useful discovery while treating agent actions and model training as different jobs.

WHAT IS THE PURPOSE?

ACT

Choose the right outcome

Allow, block, use Redirects for AI Training, deploy AI Labyrinth, or offer Pay Per Crawl and x402 access.

ALLOW · BLOCK · DIRECT · CHARGE

SUPPORTING SIGNALS

Content Signals publish intent; Web Bot Auth verifies legitimate bots; agent-readiness insights and the Analytics API provide evidence. Pay Per Crawl remains a private-beta capability.

02 · Protect the AI endpoint

Turn prompt risk into application security policy.

AI Security for Apps supplies LLM-specific context. Your WAF custom rules and Rate limiting decide what happens next.

Open Security Settings ↗Security Analytics ↗

IDENTIFY

Mark the prompt endpoint

Save the operation in Web Assets and apply the managed cf-llm label.

WHERE DO PROMPTS ENTER?

DETECT

Understand the risk

Detect Prompt injection, PII in prompts, Unsafe topics, and organization-specific Custom topics.

WHAT DID CLOUDFLARE SEE?

ENFORCE + TUNE

Observe before you block

Validate sampled traffic, then apply a narrow WAF rule or rate limit and tune it with analytics evidence.

WHAT SHOULD POLICY DO?

Four ways Cloudflare detects prompt risk

PROMPT → MANAGED DETECTORS → cf.llm FIELDS → WAF → APPLICATION MODEL

Injection scoringLower score · higher risk

METHODPattern risk

OUTPUT1–99 risk score

Fuzzy PIIEntities inside natural language

METHODEntity recognition

OUTPUTMatched PII categories

Unsafe taxonomyBuilt-in safety categories

METHODFixed S1–S14 taxonomy

OUTPUTMatched categories

Zero-shot topicsSemantic match · No custom training

METHODIntent comparison

OUTPUTScore per custom topic

cf.llm.prompt.injection_scorecf.llm.prompt.pii_categoriescf.llm.prompt.unsafe_topic_categoriescf.llm.prompt.custom_topic_categories

API Shield

>50% of Internet traffic are API calls — and that's where the breach starts.

APIs run your revenue: checkout, login, payments, partner integrations, mobile apps. They're also the shortest path to your data — machine-readable, often undocumented, and rarely covered by the same controls as your website. API Shield protects the surface your business actually runs on: it discovers every endpoint you have, validates every request that comes in, and stops sensitive data from leaking out — without a single change to your code.

Open dashboard ↗

Discover · Validate · Scan illustrative · request in, response out

API Discovery

Find what you didn't know you had.

🔍 Shadow APIs auto-catalogued Shadow & undocumented APIs are the #1 root cause of API breaches (OWASP API Top 10)

🔓 Flag publicly-exposed internal APIs Admin panels, debug endpoints, internal tools — discovered before they're exploited

Schema Validation

Reject what doesn't fit.

📐 Block requests that don't match your OpenAPI / GraphQL schema Stops Broken Object Property Authorization (OWASP API #3) — the spec you already have becomes an edge-enforced firewall

🚫 Block extra or missing fields Stops mass-assignment attacks · zero origin code changes

Sensitive Data Detection

Stop the leak before it lands.

💳 Credit card numbers (PCI) PCI-DSS non-compliance: $5K–$100K/month in fines per merchant (Visa)

🔑 API keys / secrets in responses Avg cost when secrets leak: $4.88M per incident (IBM 2024)

Edge-enforced

schema, auth, and rate checks run at the edge — zero changes to your app

Bring your spec

OpenAPI / GraphQL schemas you already have

Compliance

PCI · HIPAA · GDPR signal coverage out of the box

API Shield docs

▸ There's more

Not every rule blocks. Some shape the traffic.

Beyond Security Rules, Cloudflare's Rules family redirects URLs, rewrites paths, modifies headers, sets cache, and runs edge snippets — no app deploy required.

See it live → Rules docs ↗

Connect, protect, perform — without compromise.

When the flood comes, you're already behind the dam.

Defense in depth — managed by Cloudflare, customized by you.

We write them. We update them. You stay protected.

Your rules. ML signals included.

Count what matters. Stop abuse at the edge.

Stop the leak before it lands.

Automation isn't one bucket anymore.

Every request scored 1–99.

Catch what static signals miss.

Allow the good ones. Block the fakes.

Classify AI crawlers by intent. Act on each.

Protect what AI takes. Protect what users send.

Know which AI crawler arrived—and what job it came to do.

See demand and value

Separate Search, Agent, and Training

Choose the right outcome

Turn prompt risk into application security policy.

Mark the prompt endpoint

Understand the risk

Observe before you block

Four ways Cloudflare detects prompt risk

>50% of Internet traffic are API calls — and that's where the breach starts.

Find what you didn't know you had.

Reject what doesn't fit.

Stop the leak before it lands.

Cyber security is a board-level conversation.
We keep it a quiet one.

Connect, protect, perform — without compromise.

When the flood comes, you're already behind the dam.

Defense in depth — managed by Cloudflare, customized by you.

We write them. We update them. You stay protected.

Your rules. ML signals included.

Count what matters. Stop abuse at the edge.

Stop the leak before it lands.

Automation isn't one bucket anymore.

Every request scored 1–99.

Catch what static signals miss.

Allow the good ones. Block the fakes.

Classify AI crawlers by intent. Act on each.

Protect what AI takes. Protect what users send.

Know which AI crawler arrived—and what job it came to do.

See demand and value

Separate Search, Agent, and Training

Choose the right outcome

Turn prompt risk into application security policy.

Mark the prompt endpoint

Understand the risk

Observe before you block

Four ways Cloudflare detects prompt risk

>50% of Internet traffic are API calls — and that's where the breach starts.

Find what you didn't know you had.

Reject what doesn't fit.

Stop the leak before it lands.

Cyber security is a board-level conversation.We keep it a quiet one.

Cyber security is a board-level conversation.
We keep it a quiet one.