Security

What we do to protect your data, what our governance controls actually enforce, how we respond to a breach, and what we do not claim. Written for British Columbia, and to PIPEDA's safeguards principle.
Effective date[ EFFECTIVE DATE ]Statement of safeguards · PIPEDA Principle 7 · BC PIPA
01

Our security posture

ATORA is a governed intelligence platform. Security is not a layer added at the end of the build; the same controls that make an answer auditable — recorded evidence, recorded decisions, refusal when the basis is missing — are the controls that make the platform defensible. This page describes the safeguards actually in place, and is written as our statement of safeguards under PIPEDA Principle 7 (Safeguards) and the corresponding obligation under the British Columbia Personal Information Protection Act (BC PIPA).

Where a safeguard is aspirational rather than implemented, we say so. The section titled “What we do not claim” exists for that reason.

02

Data protection

Authentication and data storage run on a managed database service under contract to ATORA. Data is encrypted in transit using TLS, and encrypted at rest by the storage layer.

  • Account credentials are hashed with Argon2id. We never store, log or transmit a plaintext credential, and we cannot recover one on your behalf.
  • Files you upload for analysis are stored in access-controlled buckets and are reachable only through short-lived signed URLs issued to an authorised session.
  • Secrets — provider API keys, service credentials — are never stored in the source repository. They are injected into the runtime environment and read only inside server-side handlers.
  • Server-side code paths that carry elevated privilege are separated from code that can reach the browser, so privileged credentials cannot be bundled into a client build.
03

Access control

  • Every database table carrying user or platform data has row-level security enabled, with policies scoped to the authenticated identity. Access is denied by default; a policy grants it explicitly or not at all.
  • Authority is never inferred from client-side state. Roles and Founder authority are resolved server-side from a governed registry, never from local storage, a profile field or a request header.
  • A second authentication factor and WebAuthn are supported on privileged surfaces.
  • A password-recovery session is a restricted class of session: it may only reach the password reset screen and cannot be used to read data or perform operations.
04

Infrastructure

The application is hosted on managed infrastructure with server-side logic executed in an isolated runtime. There are no long-lived operator shells in the production path, and the platform exposes no interface that grants arbitrary command execution against production systems.

  • Operational tooling that reaches a Founder's own machine does so through an outbound-only bridge that the Founder starts and stops. It never accepts inbound connections and it cannot reach the certified core or any secret.
  • Public endpoints intended for external callers are separated from authenticated application logic, and each verifies its caller inside the handler.
  • Traffic is served over HTTPS only, canonicalised to a single production origin.
05

Governance controls

These are the controls that distinguish ATORA from a conventional application, and they are security controls as much as governance controls.

  • Append-only audit. Audit events are append-only and the constraint is enforced at the database level by triggers that reject updates and deletes. An audit record cannot be quietly rewritten, including by us.
  • Risk-classified privileged actions. A privileged action is classified by risk and routed accordingly — automatic execution, execution with audit, human review before execution, or refusal. Classification escalates; it never de-escalates on the basis of a request.
  • Sovereign control that fails closed. A sovereign control can suspend autonomous operation across the platform. If the control store cannot be read, autonomy is denied rather than granted. An unknown state is treated as a stop, not a permission.
  • Cryptographic core manifest. The certified core is protected by a manifest of file digests that is verified on every certification run. Any modification to a frozen file is detected and fails the run.
  • No model authority. A language model never executes an action, grants an authority, or produces a probability, forecast or confidence figure. Those come only from the platform's certified path.
06

Breach response and notification

If a breach of security safeguards occurs, we contain it, assess the risk of harm, record it, and notify as the law requires. Our obligations, stated explicitly:

  • Under PIPEDA, a breach of security safeguards that creates a real risk of significant harm to an individual is reported to the Office of the Privacy Commissioner of Canada and notified to affected individuals as soon as feasible. Other organisations and institutions are notified where doing so may reduce the risk of harm.
  • Under BC PIPA, notification to affected individuals is mandatory where the same real-risk-of-significant-harm threshold is met. Reporting the breach to the Office of the Information and Privacy Commissioner for British Columbia (OIPC BC) is currently voluntary. We report to the OIPC BC regardless. The stricter federal standard already applies to our cross-border activity, and we do not intend to operate two different thresholds depending on which statute a given record falls under.
  • We maintain a record of every breach of security safeguards for 24 months, whether or not it was reportable, and make those records available to the regulator on request, as PIPEDA requires.
  • Where the personal information of Quebec residents is involved, we maintain a register of confidentiality incidents and notify the Commission d'accès à l'information du Québec as Quebec Law 25 requires.

A notification will describe what happened, the personal information involved, what we have done, what you can do to reduce your risk, and how to reach our Privacy Officer.

07

Vulnerability disclosure

We welcome good-faith security research. Report a suspected vulnerability to hello@atora.ai with enough detail to reproduce it.

  • We acknowledge reports within [ RESPONSE WINDOW ].
  • We ask for reasonable time to remediate before public disclosure, and we will keep you informed of progress.
  • We will not pursue legal action against good-faith research that respects the privacy of others, avoids service degradation, does not exfiltrate or destroy data, and does not use a finding for any purpose beyond demonstrating it.
  • Please do not test against other users' accounts or data. If you need a test account, ask us for one.

We do not currently operate a paid bug bounty.

08

What we do not claim

We would rather be believed than impressive. So, plainly: we do not currently hold SOC 2, ISO 27001 or any equivalent third-party security certification, and nothing on this site should be read as implying otherwise. Our current status is [ CERTIFICATION STATUS ].

Our internal certification programme — the golden-baseline manifests and conformance suites referenced above — is our own engineering discipline. It is real, and it is verifiable, but it is not an external audit and we do not present it as one.

We also do not claim that any system is unbreachable. We claim that our safeguards are proportionate, that our audit trail is tamper-evident, and that if something goes wrong you will hear it from us.

Questions about this page? hello@atora.ai