Security posture, stated plainly.
Everything below is a statement about our codebase and infrastructure, with a claims-ledger id linking it to the code that makes it true — the same standard as every other page here. The last section lists what does not exist yet, because a posture you can verify includes the gaps.
Safe by construction, where it matters most
The strongest control is a state that cannot be represented. Where we could build the risk out of the data model, we did.
Your broker password has nowhere to live
The broker-session model has no field a password could occupy, and a scanner refuses any payload that carries one. We integrate with brokers through their token flows; the credential itself never reaches us. (CL-009)
Broker tokens are envelope-encrypted
Access tokens are encrypted with AES-256-GCM under a Key Vault key, bound to their own database row so a ciphertext cannot be replayed into another account, with key rotation designed in. Production refuses to start with the development cipher. (CL-031)
The sandbox is virtual money, full stop
Paper trading uses virtual money on live prices. A real order can only leave through four gates — deployment flag, versioned consent, fresh second factor, and your explicit confirmation on every single order. (CL-009)
Where your data lives, and why it stays there
Akechi Trade is built for the DPDP Act 2023 posture: Indian-region processing, purpose-limited consent, no dark patterns.
India-only, checked by machines
Our infrastructure code is linted by 52 checks that refuse non-India regions and public network access before anything deploys — the policy is a build failure, not a memo. (CL-013)
Consent is per purpose, and versioned
A contact you verify for alerts is used for alerts. Consent for live trading binds to the exact wording you saw — if the wording changes, the old consent is invalid and you are asked again. (CL-009)
Leaderboard consent works in both directions
The leaderboard is opt-in and pseudonymous, and consent is re-checked when the board is served: withdraw it and the very next read excludes you. (CL-023)
Sessions, staff and the audit trail
Operational controls, as implemented — not as aspired to.
Sessions expire at the server
Thirty minutes idle and the session is revoked server-side. A stale tab does not stay signed in because its JavaScript stopped running. (CL-029)
Staff are not users
Staff authenticate against a separate directory from customers, and roles are split by consequence: the role that approves research publication is not the role that operates the platform. In production, a missing staff configuration refuses access rather than degrading. (CL-030)
Four audit logs no one can rewrite
Research, risk, billing and AI chat each write hash-chained, append-only records, and every read re-verifies the chain. Corrections attach; history does not change. (CL-017)
What doesn't exist yet
No external security audit yet
Akechi Trade has not yet been through an independent penetration test or security audit. One is planned before general availability; when it has happened, this page will say so — and until it has, this page will not pretend otherwise.
Dual control on staff actions is designed, not implemented
Destructive staff operations are role-separated today; requiring a second approver is on the roadmap. We say this because a security page that lists only finished things is an advertisement, not a posture.
Responsible disclosure, pre-launch
A security.txt and dedicated disclosure address ship with the public launch. Until then, security reports reach the same team through the waitlist contact path — small platform, short path.
How this posture is enforced — the CI gates, the invariants, the full receipts table — is on the engineering page. Data rights and retention are in the privacy notice.