Security & compliance
Built like it holds your keys.
Because it does.
A platform that connects to brokerage accounts owes its members more than a promises page. Here is how the system is actually built, what we refuse to touch, and where the compliance program stands — stated plainly, including the parts that are still in progress.
Practices
The controls in place today.
Encryption everywhere
TLS in transit, encryption at rest across the stack. Brokerage OAuth tokens get a second layer of application-level encryption, with keys held in Azure Key Vault — never in code or config.
Your broker credentials never touch us
The brokerage connection uses the broker's own OAuth flow. We never ask for, see, store, or automate entry of your username, password, or two-factor codes — and we never will. Disconnecting deletes the tokens immediately.
No order without your authority
The platform cannot place, modify, or cancel a live order on its own initiative — architecturally, not just by policy. AI agents stop at drafts; live order actions run only under authority you explicitly granted yourself.
Agent code runs in a hard sandbox
Analytics the AI writes execute in an isolated process with no network access, no environment, strict resource limits, and an import allowlist. Experimental code can never touch canonical records.
Least-privilege access
Role-based access control at every layer, managed identities instead of shared secrets in our cloud, audit logging on sensitive operations, and per-tenant data scoping enforced in the data layer.
Tested against production-shaped data
Non-production environments never see raw production data. A one-way sanitization pipeline pseudonymizes identities and strips personal content before anything leaves production, on a default-deny field policy.
Devices you trust, sign-ins that step up
The platform remembers the devices you've verified. A sign-in from anywhere else triggers extra verification before it gets in — and you can review and revoke trusted devices at any time.
One person, one account
Duplicate signups are refused, so a banned member can't walk back in under a new name — and every track record belongs to exactly one accountable human.
Private chat stays private
One-on-one and small-group chats are end-to-end encrypted: messages are sealed on your device and only your devices hold the keys. Our servers relay them but cannot read them.
Custom code watched for known flaws
Every Protocol — the custom analytics and programs the agent builds — is checked against public vulnerability databases, and an optional dial pauses a Protocol automatically if a critical flaw is disclosed in something it uses.
Infrastructure
Boring on purpose.
The stack is built on providers whose security programs are audited at a depth no startup can match — and the sensitive parts are delegated to them by design: identity to Microsoft, card data to Stripe, key custody to Azure Key Vault.
Isolation scales with the plan: members on the Experiment tier and above get their own dedicated database, Enterprise and White Label organizations run on fully dedicated infrastructure, and organizations can raise internal data walls between their own units.
Compliance posture
What we are, and what we aren't.
- Analytics, not advice. Thesis Trading AI is analytical tooling. We are not a registered investment adviser, broker-dealer, or commodity trading advisor, and the platform does not recommend securities or place orders on its own. See the Investment Disclaimer.
- US-only, by design. The service is offered to US residents 18 and older, and our privacy program is built to US law — CCPA/CPRA and comparable state statutes. See the Privacy Policy.
- Research data, disclosed. The aggregated research datasets we build are anonymized with minimum-population thresholds, exclude Enterprise and White Label organizations entirely, and are disclosed plainly in the Privacy Policy — not buried in it.
- SOC 2 criteria, honestly stated. Controls are built to SOC 2 criteria from the start — access control, change management, audit logging, vendor management — and a formal SOC 2 Type II examination follows as the platform scales. The report status is published here; we don't claim the badge before the report exists.
Found something?
Report it. A person reads it.
Security reports go to Security@ThesisTrading.AI and are read personally — machine-readable details live in security.txt. Good-faith research is welcome; we ask only that you avoid accessing other members' data and give us reasonable time to fix what you find.