Security
Atlas Forms includes several security layers to protect your forms from abuse: origin validation, sandbox mode for local testing, rate limiting, a honeypot field, attack-pattern detection with spam quarantine, automatic IP blocking, optional Cloudflare Turnstile, and an optional AI spam filter.
Origin Validation
Restrict which domains can submit to your forms.
Configuration
example.com, app.example.com)How It Works
Origin or Referer headerexample.com allows www.example.com)Localhost Testing
Localhost submissions are accepted only while a form is in sandbox mode. Until you register an IP address, any localhost origin can submit to a sandbox form. Once at least one IP is registered, only registered IPs can.
Why IP Registration?
Localhost origins (http://localhost:3000) are identical for all developers. Without IP validation, anyone who discovers your form ID could submit to your form from their localhost.
By registering your IP, Atlas ensures only your machine can submit to the form during development.
How to Test Locally
Form in SANDBOX mode, no IPs registered:
- Any localhost → ✅ Allowed
Form in SANDBOX mode, IPs registered:
- Your localhost + registered IP → ✅ Allowed
- Other localhost + different IP → ❌ Rejected
Form in LIVE mode:
- Localhost (any IP) → ❌ BlockedIP Registration
Your public IP (detected via Cloudflare) is shown in form settings. Click Register My IP to authorize it.
Common scenarios:
| Scenario | Solution |
|---|---|
| IP changed (new WiFi, VPN) | Click "Register My IP" again |
| Several developers | Each developer registers their own IP |
| Same office (shared NAT) | Same IP = same org, works automatically |
Optional: Sandbox API Key
For extra security in JavaScript apps, generate a Sandbox API Key:
.env.local:ATLAS_FORM_KEY=sk_sandbox_xyz123...await fetch('https://atlasforms.app/f/abc123', {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'X-Atlas-Key': process.env.ATLAS_FORM_KEY
},
body: JSON.stringify({ name: 'Test' })
})This provides two layers of security: IP binding + API key.
When You Go Live
When you switch from Sandbox to Live mode:
This ensures your production form only accepts submissions from real users on real domains.
Error Response
{
"error": "Origin not allowed"
}Rate Limiting
Prevent abuse with per-IP rate limiting.
Limits
| Context | Limit |
|---|---|
| Form submissions | 60/minute per IP, per project |
API reads (GET) | 100/minute per API key or account token |
API writes (POST, PATCH, DELETE) | 50/minute per API key or account token |
These limits are fixed; there is no setting to change them.
Rate Limit Headers
X-RateLimit-Limit: 60
X-RateLimit-Remaining: 55
X-RateLimit-Reset: 2024-01-15T10:31:00ZWhen Limited
{
"error": "Rate limit exceeded. Please try again later.",
"retry_after": 45
}Status code: 429 Too Many Requests
Spam Protection
Honeypot Fields
Add invisible fields that only bots fill out:
<form action="https://atlasforms.app/f/abc123xyz" method="POST">
<!-- Honeypot - hidden from humans -->
<div style="position: absolute; left: -9999px;">
<input type="text" name="_gotcha" tabindex="-1" autocomplete="off">
</div>
<input type="text" name="name" required>
<button type="submit">Send</button>
</form>If _gotcha or _honeypot has a value, the endpoint returns a normal success response but discards the submission. It is not stored and does not count toward your monthly quota.
Turnstile CAPTCHA
Cloudflare Turnstile is an optional CAPTCHA check, off by default:
When it is on, browser form posts without a valid cf-turnstile-response token are rejected. JSON requests are not checked.
<form action="https://atlasforms.app/f/abc123xyz" method="POST">
<input type="text" name="name" required>
<div class="cf-turnstile" data-sitekey="0x4AAAAAAA..."></div>
<button type="submit">Send</button>
</form>
<script src="https://challenges.cloudflare.com/turnstile/v0/api.js" async defer></script>Spam Quarantine
Public form endpoints attract automated vulnerability scanners. A submission
that matches a known scanner fingerprint — sqlmap's EXTRACTVALUE(...) and
UNION ALL SELECT ... NULL-- probes, injected
— is quarantined instead of delivered:
Nothing is thrown away, because detection is not perfect. If a real message
lands there, select it and choose Not spam — it moves back to your
submissions, the notification email is sent, and it counts as a normal
submission from then on. You can also mark an ordinary submission as spam,
which refunds its quota slot.
The scanner sees exactly the same response as a delivered submission, so it
gets no signal about what was detected.
Quarantine is on by default and can be switched off per form under
Form Settings → Security → Spam Quarantine. With it off, every probe
is emailed to you and consumes quota.
AI Spam Filter
The attack-pattern detector catches scanners, not ordinary spam such as SEO
pitches, link drops or crypto offers. For those, you can turn on the
AI spam filter per form under Form Settings → Security. It is free on
every plan and off by default.
When it is on, the text of each submission (field names and values, never
uploaded files) is sent to Cloudflare Workers AI
and classified by Cloudflare's Clef-flash decision model. A submission it
classifies as spam is quarantined exactly like an attack probe: stored in the
Spam tab with a reason such as *AI filter: sales pitch*, with no email,
webhook or integration delivery, and no charge against your quota. The
submitter gets the normal success response.
The filter errs toward delivery. If the model is slow (over about 1.5 seconds)
or unavailable, the submission is delivered as normal. A real message that is
misclassified can be restored with Not spam.
You can also set it from the API, CLI or MCP server: ai_spam_filter when
creating a form (POST /api/v1/projects/{id}/forms, atlas forms create --spam-filter,
atlas_create_form) or on an existing form (PATCH /api/v1/forms/{id},
atlas forms update --spam-filter on|off, atlas_update_form).
Spam in the API
GET /api/v1/submissions leaves quarantined submissions out unless you pass
include_spam=true. Every item has is_spam and spam_reason. See the
IP Blocking
Atlas blocks IP addresses that repeatedly abuse form endpoints.
Automatic Blocking
Each of these counts as a violation for the sending IP:
After 5 violations, the IP is blocked from all Atlas endpoints for 24 hours. Blocks expire automatically.
Manual Blocking
Atlas staff can also block or unblock an IP. There is no customer-facing IP blocklist.
Webhook Security
Signature Verification
When the project has a webhook secret, the first delivery attempt includes an HMAC-SHA256 signature of the raw request body:
X-Webhook-Signature: sha256=abc123...Verify it to reject spoofed requests:
const crypto = require('crypto');
function verifySignature(payload, signature, secret) {
const expected = 'sha256=' + crypto
.createHmac('sha256', secret)
.update(payload)
.digest('hex');
return crypto.timingSafeEqual(
Buffer.from(signature),
Buffer.from(expected)
);
}Webhook Secrets
The dashboard and API do not currently let you set or view a webhook secret, so deliveries are not signed with a secret you control. Until that is available, use an unguessable webhook URL (for example, one that includes a random token) and check X-Submission-Id against the API if you need to confirm a delivery. See Webhooks for details.
URL Restrictions
To prevent server-side request forgery, webhook URLs in live mode must use HTTPS and cannot point at localhost, private IP ranges, or cloud metadata endpoints. Sandbox mode allows localhost, private IPs, and HTTP for local testing. Metadata endpoints are always blocked.
CSRF Protection
The dashboard uses CSRF tokens to prevent cross-site request forgery.
For API integrations:
Data Security
Encryption and Storage
Data Retention
Data Export and Deletion
GET /api/v1/submissions?format=csv)Security Headers
Atlas sets these security headers on its web pages and form endpoints:
X-Content-Type-Options: nosniff
X-Frame-Options: DENY
X-XSS-Protection: 1; mode=block
Referrer-Policy: strict-origin-when-cross-origin
Strict-Transport-Security: max-age=31536000; includeSubDomains; preload
Permissions-Policy: camera=(), microphone=(), geolocation=(), payment=(self)
Content-Security-Policy: default-src 'self'; ...Reporting Vulnerabilities
Found a security issue? Email security@atlasforms.app.
We appreciate responsible disclosure and will: