Security

How your screenplay is protected

An unreleased script is the most sensitive document in this industry. Here is what protects yours, control by control — and, just as plainly, what we have not certified yet. Last updated August 28, 2026 · ScriptRisk is in beta.

IN TRANSIT your script TLS — encrypted tunnel browser → us, and to every provider ScriptRisk AT REST — TWO LAYERS Google Cloud Storage encryption application-layer seal — Fernet (AES + HMAC) key in Secret Manager

Bucket access alone yields ciphertext: the seal is applied before your script reaches storage, and the key never lives beside the data.

Encryption

The moment your file arrives, before it touches any disk, it is sealed with an authenticated cipher (Fernet: AES-128-CBC + HMAC). Only then does it enter Google Cloud Storage — which wraps it in a second, independent layer of encryption. Two locks, two different keys, held in two different places.

This is what someone with full access to our storage bucket — a stolen credential, a misconfigured permission, a subpoenaed disk — actually sees of your scene:

records/eb6…f8.json →
gAAAAABqkT6ZB1YUMTMQZF4echbpf7MsXXu4YzXLph0-GX9A-0dLjVEf0cvuc-l5NVZi_limT84rQC2Nx73mskUGRIoYp1TYAfPwtzL8VNVxIRLTZMtVe-LI…

Not your logline. Not a character name. Not one word of dialogue. Ciphertext is all the bucket ever holds — the key lives apart, in Google Secret Manager, never in code, never in configuration, rotated on credential-exposure events and major releases (last rotation: August 2026).

Access

We never see a password, because there are none. Sign-in is Google OAuth only; sessions expire on their own.

Ownership is not a policy we promise to follow — it is the shape of the storage itself. Your analyses live under your key prefix:

users/《your-account》/runs/a1d5…d9.json   ← reachable only through your session
users/《someone-else》/runs/…          ← no code path leads here from yours

Reading another writer's run is structurally impossible, not merely forbidden. Anonymous runs are guarded by unguessable identifiers — the link is the key, and only you hold it. Production access is limited to a single administrator; hardening of every provider account (hardware-backed two-factor, access reviews) is part of the TPN preparation below.

Your work is never training data

We do not train models on your screenplay. Ever. Our AI runs on Google Cloud Vertex AI under enterprise terms that forbid training on your content — the same terms studios rely on.

And our own operational logs are engineered to be script-blind. A completed analysis writes a line like this — and only like this:

run_completed  flags=21  rejected=3  elapsed=297s  desks=4

No title. No character names. No excerpt. Not one word of your screenplay is ever written to a log — the privacy page makes that promise, and the logging code enforces it.

Deletion that actually deletes

One click in My reports (for anonymous runs, contact us), and every artifact of that analysis is destroyed together — not hidden, not unlisted, removed:

✕ scripts/…fountain      the screenplay itself
✕ records/…json        the full analysis
✕ pdfs/…binder.pdf     the clearance binder
✕ pdfs/…onesheet.pdf   the one-sheet
✕ owners/…             the ownership marker

The report page 404s. The binder link dies. There is no recycle bin we quietly keep.

Monitoring and response

A paid analysis failing silently is treated as an incident, not a statistic: any run failure pages the operator by email within five minutes, around the clock.

And nothing reaches production on hope. Every deploy must survive the full test suite, lint, a contest-compliance scan, and a live-traffic check — mechanically:

$ safe_deploy
  128 tests … green · lint … clean · compliance … OK
  1 analysis in flight — refusing to deploy.

Yes — the deploy tool refuses us while your analysis is running. Your run finishes on the code it started with. The platform beneath is Google Cloud, inheriting Google's own SOC 2, ISO 27001, and related attestations for the physical and infrastructure layers.

What we have not certified — yet

Badges mean something only if you know what they prove. So, plainly:

  • TPN (Trusted Partner Network) — the MPA's own content-security program for vendors handling unreleased material. Membership and the Blue Shield assessment are next on our roadmap after the current release cycle.
  • SOC 2 — an independent, months-long audit of the controls this page describes. Planned as we grow into studio and enterprise work.

We will not display a badge we have not earned. When these change, this page changes.

Questions or reports

Security questions, or something you believe we should fix: contact us. Security reports are read first and answered first.