Skip to content
Intuist

SECURITY

SOC 2 TYPE II · CASA · REPORT UNDER NDA

What we check, and what we do not claim.

Security here is not a badge wall. It is a list of things that run on every build or when you publish from the builder, a list of certifications we actually hold, and — the part most vendors leave off — a list of the ones we do not.

01 · VERIFICATIONS

17

automated checks on every build and refine

Intuist platform, September 2026

Six layers: five on every build, the red team on publish from the builder.

Not a quarterly checklist. Five run with every build, the checks grading it as it finishes, and the red team runs when you publish from the builder.

0102030405
A SECTION THROUGH ONE BUILD · YOUR MESSAGE ENTERS AT 01 AND THE LIVE APP LEAVES AFTER 05 · 03 GRADES THE WORK AND REPORTS EVERY RESULT; IT DOES NOT HOLD THE BUILD BACK · THE RED TEAM IS NOT IN THIS PATH: IT RUNS WHEN YOU PUBLISH FROM THE BUILDER
INPUT GUARD
Every user input is scanned for prompt injection, PII leakage, and adversarial patterns before it reaches any AI model.
SANDBOXED EXECUTION
Every app's code runs in its own read-only container, with its memory, CPU and run time capped. It cannot open a connection to our private network or the cloud's metadata server.Measured from inside a container, 9 October 2026 · the public internet stays open, plate 04
17 AUTOMATED CHECKS
Every build and every refine streams 17 checks: 6 security, 8 quality, 3 performance.
ENCRYPTION IN TRANSIT AND AT REST
Data encrypted in transit (TLS 1.2 and 1.3) and at rest (AES-256). Platform secrets live in Google Secret Manager.
ZERO-TRUST ACCESS
Email-verified passkey authentication with httpOnly cookies. No tokens in localStorage. Domain and email-level access control.
RED-TEAM TESTING
When you publish an app from the builder, a red-team scan runs in the background: adversarial prompts against its AI assistant (injection, jailbreaks, data extraction) and a review of its code for XSS and unsafe patterns.

EACH ROW RUNS AUTOMATICALLY · NO ONE HAS TO REMEMBER IT

SECURITY REVIEW AS AN ENGAGEMENT
Pen-test style review and fixes for your own apps, sites, systems and clusters, scoped and delivered as an engagement.Intuist platform, September 2026

SCOPED SEPARATELY · NOT PART OF THE AUTOMATIC STREAM

02 · THE SEQUENCE

From the message to the live app.

Five stations, in order. None is optional; the red team waits for you to publish from the builder, and the rest need no one to press anything.

  1. 01

    GUARD

    Input Guard scans every message for injection, PII and adversarial patterns before the model sees it.

  2. 02

    SANDBOX

    The code runs in its own read-only container, refused our private network.

  3. 03

    CHECK

    Seventeen automated checks grade the code: transport, secrets, injection sinks, access control, accessibility and more.

  4. 04

    RED TEAM

    When you publish from the builder, adversarial prompts and a code review probe the app for injection, jailbreaks and XSS.

  5. 05

    WATCH

    In production, runtime errors are diagnosed and patched without waiting for a person.

03 · CERTIFICATIONS

What we hold, and where to check it.

Every row below is auditable. The SOC 2 Type II report is available under NDA through the Trust Center.

SOC 2 TYPE II
An independent auditor tested our security controls from 9 October 2025 to 9 January 2026: unqualified opinion, no exceptions.Security criteria · report dated 11 March 2026, available under NDA · trust.intuist.ai
CASA
Cloud Application Security Assessment, Tier 2: assessed by TAC Security, valid to 11 December 2026.Letter of validation issued 10 December 2025 · intuist.ai/casa-labs-certificate.pdf
GOOGLE CLOUD PLATFORM
Deployed on GCP Cloud Run with auto-scaling, regional redundancy, and Google's enterprise-grade physical security.
OWASP
Every generated app is automatically audited against the OWASP Top 10 — injection, XSS, CSRF, and more.
BAA
Business Associate Agreements available for healthcare clients handling PHI under HIPAA requirements.Our own BAAs with Google Cloud, Google Workspace, OpenAI and MongoDB Atlas are executed · September 2026

HELD TODAY · OCTOBER 2026

View our Trust Center

04 · NOT CLAIMED

The things we will not let you assume.

A vendor who lists only what they hold is asking you to infer the rest. Here is the rest, in the same type size.

“A trust score assesses the code Veda generated. It does not audit your deployment, your staff or how you handle your data. No other AI builder ships a verdict with its output at all.”

INTUIST AI · THE CAVEAT WE PRINT OURSELVES

GDPR AND CCPA
We do not claim GDPR or CCPA compliance, and we would rather tell you that than let you assume it.Full wording on /help
HEALTH PRIVACY
We do not claim HIPAA compliance. What is true today: our BAAs with Google Cloud, Google Workspace, OpenAI and MongoDB Atlas are executed, and we sign one with healthcare clients.Ask sales@intuist.ai for the agreement
YOUR DEPLOYMENT
A trust score assesses the code Veda generated. It does not audit your deployment, your staff, or how you handle your data.
AN OFFLINE SANDBOX
We do not claim an app's code is cut off from the internet. It can reach the public internet, because apps call the services you connect. Our private network and the cloud's metadata server are refused.Sandbox egress guard · measured 9 October 2026
DATA RIGHTS, IN PRACTICE
Access, export and deletion of your data on request, honoured within 30 days, across every system we store it in.Email privacy@intuist.ai

STATED PLAINLY · FULL WORDING ON /HELP

05 · ARCHITECTURE

The boundaries your data flows through.

Three columns, no diagram. A hub-and-spoke drawing of a network is decoration; the boundary names are the fact.

SANDBOX VMS
VPC-isolated
CODE EXECUTION
Private network refused
TRANSPORT
TLS everywhere
EDGE
HTTPS only, TLS 1.2 and 1.3

NETWORK ISOLATION

AT REST
AES-256
SECRETS
Google Secret Manager
LOGS
No PHI
PII
Automatic redaction

DATA PROTECTION

PROMPT INJECTION
Detected before the model
ABUSE
Rate limiting and throttling
BEHAVIOUR
Anomaly detection
COVERAGE
Monitored around the clock

THREAT PREVENTION

CHAT AND DOCUMENTS
Anthropic's Claude models, run by Anthropic and by Amazon Web Services (Bedrock).
VOICE
ElevenLabs (in the Veda app, OpenAI when that road is unavailable), streamed from your device; the audio does not pass through our servers.
BUILDING
A coding model we run ourselves, then Meta (Muse Spark), then Anthropic's Claude models or Google as fallbacks.
INSIDE THE APPS
The AI model each app is set up to use: Anthropic's Claude models, run by Anthropic and by Amazon Web Services (Bedrock); OpenAI; Google (Gemini); or a model Intuist runs itself.Full wording, and each provider's terms on training, in clause 11 of intuist.ai/privacy

AI PROCESSORS

WEBSITE FORMS
Cloudflare Turnstile, an invisible human check on the brief, contact and partner forms on intuist.ai. It reads browser and device signals and the visitor's IP address to tell a person from a bot.Clause 8 of intuist.ai/privacy, and Cloudflare's Turnstile Privacy Addendum

WEBSITE PROCESSORS

veda@intuist.aiRequest a security review

Ask for the SOC 2 report, a specific control, or a straight answer about an obligation we have not listed. You will get one either way.