Trust the math. Verify us, don’t take our word.

Most vendors ask you to trust their operational controls. We ask you to trust arithmetic. Every AI-assisted event inside a manuscript — the prompt you sent, the response returned, each accept, edit, or dismiss you made against a Redline note — is hashed and cryptographically linked to the previous event.

The result is an Authorship Packet: an ordered chain that a third party can verify offline without connecting to our servers, our database, or anything we control. If a single character of the record is altered, the chain breaks and any reviewer can see it.

This is why we don’t lead with a compliance mark. A badge attests that a firm reviewed our controls at a point in time. A hash chain attests to the record itself, at the moment of writing, verifiable by anyone, forever.

All traffic between your browser and RedInkAI is served over TLS 1.2+. Manuscript content, InkTrail events, and account records live in MongoDB Atlas with AES-256 encryption at rest. Backups inherit the same encryption posture; nothing is written to disk unencrypted at any point in the pipeline.

Passwords are hashed with bcrypt using a per-account salt. Session tokens are short-lived JSON Web Tokens signed with a secret rotated on server restart; they never leave the browser’s local storage boundary.

Access to production data is limited to the founder. There is no support team with read access. There is no engineering team with read access. When we need to reproduce a specific bug you have reported, we ask for your permission, in writing, before opening any part of your project.

Redline’s proactive scan runs only when your session is open in an editor tab. Closing the tab stops the model calls; there is no server-side batch that reads your manuscript in the background.

When you enable Trail Lock on a project, the InkTrail becomes append-only from the platform’s side. Even the founder cannot rewrite a locked chain; every audited operation is logged to a separate immutable event stream.

We retain your manuscripts, InkTrail events, and account metadata for as long as your account is active. If you delete a project, its content is removed from the primary datastore within 30 days and expires from encrypted backups within a further 60 days (90 days total).

If you close your account, everything above happens automatically. We keep a minimal record of the deletion event (account ID, timestamp, requester) for two years to defend against fraudulent recovery attempts. That record contains no manuscript content.

Application logs (request paths, error codes, IP-hash) are retained for 30 days for operational debugging and then rotated out. We do not retain manuscript text in logs.

You can export any project as a signed Authorship Packet at any time from the editor. You can delete any project from the dashboard; the deletion cascades as described above.

For GDPR / CCPA / UK GDPR data-subject requests (access, portability, rectification, erasure, restriction), write to redinkaistudio@redinkai.com. We acknowledge within 72 hours and complete verified requests within 30 days.

If you are not the account holder but believe your personal data is in a manuscript on our platform (for example, you are named in someone’s memoir draft), write to the same address and we will work with you and the account holder toward a resolution.

Your manuscripts, prompts, Redline notes, and reference materials are never used to train RedInkAI models. We do not sell aggregated writing data. We do not share manuscript content with third parties for advertising, analytics, or benchmarking.

Inference-time calls to model providers are made under their commercial API terms, which specifically exclude training on submitted content. We select vendors that make this contractual, not aspirational.

If we detect a security incident affecting your data, we notify affected account holders by email within 72 hours of confirmed impact, disclose the scope in plain language, and publish a post-mortem within 14 days.

Report a suspected vulnerability or incident to security@redinkai.com. We commit to acknowledging security reports within 48 hours. Responsible disclosure is welcomed; we will not pursue legal action against researchers acting in good faith under the guidelines below.

• Do not access, modify, or delete data belonging to other users.
• Do not disrupt service (no DoS, no automated scanning at load).
• Give us a reasonable window to remediate before public disclosure (default: 90 days).
• Tell us what you found, how to reproduce it, and what impact you assess.

Our internal information security policies (access control, data retention, vendor management, incident response, cryptographic standards, business continuity, acceptable use) are version-controlled and reviewed annually. We do not currently hold a SOC 2 attestation; we will pursue one when a specific institutional pilot requires it, and until then we would rather be honest about that than post a badge.

For a copy of any specific policy under NDA, or for a security questionnaire response (SIG Lite, CAIQ), write to redinkaistudio@redinkai.com.

RedInkAI is designed with 21 CFR Part 11 readiness in mind — append-only audit trail, deterministic timestamping, cryptographic integrity of the record, and clear user attribution. Pilot deployments in FDA-regulated workflows are handled under a signed Business Associate or Master Services Agreement.

For legal-industry deployments, the Authorship Packet is structured to support work-product declarations and privilege claims when the record is exported and archived by the practice, not by us.