Responsible Use

What ATORA commits to, what it refuses to do, what we ask of you, and how to ask a human to review a result. Mapped to Canada's voluntary AI code, because Canada has no general-purpose AI statute.
Effective date[ EFFECTIVE DATE ]
01

Our commitments

These are not aspirations. Each one is a property of how the platform is built, enforced in code and checked by its conformance suites.

  • A language model may never author a probability, forecast or confidence figure. Those come only from the certified path. A model may put a governed result into clear words; it may not decide what the result is.
  • Every answer carries its evidence, its stated confidence and its limits. An answer without those attachments is not shipped as an answer.
  • The system declines when the evidence is not there, and a refusal names what was missing. Silence and invention are both worse than a stated refusal.
  • Reasoning uses only what was knowable at the time being reasoned about. Temporal integrity is enforced, so a result cannot be improved by information from after the fact.
  • Material actions require recorded human approval. Privileged operations are risk-classified, and classification escalates rather than relaxes.
  • We publish honest build status, including for capabilities that are specified but not yet built. Something not built is described as not built.
  • Unknown is never rendered as zero. A missing value is shown as missing, never as a number that happens to look reassuring.
02

How this maps to Canada's voluntary AI code

Canada has no general-purpose artificial intelligence statute. The federal Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems (Innovation, Science and Economic Development Canada, September 2023) sets six commitments. We map our design onto them so the claim can be checked rather than trusted.

ATORA's design mapped to the six ISED voluntary code commitments
Code commitmentHow ATORA addresses it
AccountabilityNamed Founder authority with a governed designation registry; append-only audit of every privileged action; a published certification baseline with a cryptographic file manifest.
SafetyA sovereign control can suspend autonomous operation and fails closed — if the control store cannot be read, autonomy is denied. Risk-classified actions escalate to human review or refusal.
Fairness and equityNo automated decision about an individual with legal or similarly significant effect without human review; prohibited uses include discrimination against protected groups; output is an estimate with limits, not a verdict about a person.
TransparencyEvery answer discloses its evidence, confidence and limits; provider processing is disclosed in the Privacy Policy; build status is published honestly, including for unbuilt capabilities.
Human oversight and monitoringMaterial actions require recorded human approval; a language model never executes an action or grants an authority; monitoring and criticality escalation operate independently of the model layer.
Validity and robustnessCalibration measured against realised outcomes with declared sample sizes and temporal splits; a model stays in draft until it beats a deliberately simple baseline on held-out data; conformance and failure suites run on every certification.

Signatory status: [ ISED CODE SIGNATORY STATUS ]. We state this explicitly and we do not imply signature unless it has actually been given. Mapping our practice onto the code's headings is not the same thing as signing it.

We do not cite the Artificial Intelligence and Data Act as binding law. Bill C-27 died when Parliament was prorogued on 6 January 2025 and no successor has been enacted.

03

What ATORA will not do

  • It will not produce a probability, forecast or confidence figure authored by a language model.
  • It will not present an estimate as a statement of fact about the future, or as professional advice.
  • It will not invent a value to fill a gap — no fabricated percentage, trend, progress figure or status.
  • It will not answer where governance, evidence sufficiency or scope forbid it.
  • It will not execute a privileged action on a model's initiative, or grant itself an authority it was not given.
  • It will not reach a secret, a credential or the certified core through any conversational or operational surface.
  • It will not claim a certification, an endorsement or a regulatory approval it does not hold.
  • It will not use information from after the period being reasoned about in order to look more accurate.
04

Your responsibilities

  • Treat output as an input to your judgement, never as a substitute for it, and never as the sole basis for a consequential decision.
  • Understand what calibration means: a probability is measured across many cases, and a well-calibrated estimate is still wrong sometimes.
  • Have the rights to the data you submit, and a lawful basis for any personal information in it.
  • Keep a human decision-maker in the loop wherever an outcome affects a person's rights, employment, credit, insurance, housing, health or liberty.
  • Represent ATORA's accuracy and status accurately when you cite it.
  • Respect refusals, rate limits and governance boundaries rather than routing around them.
05

Prohibited uses

You must not use ATORA to:

  • Carry out or facilitate unlawful activity.
  • Make automated decisions about individuals with legal or similarly significant effects without human review.
  • Discriminate against individuals or groups protected under the BC Human Rights Code or the Canadian Human Rights Act.
  • Conduct surveillance of individuals, or profile, locate or identify a person without a lawful basis and their knowledge.
  • Generate misleading claims about ATORA's accuracy, certifications, capabilities or endorsements.
  • Resell or expose output as regulated advice — financial, legal, medical, insurance or otherwise professional.
  • Attempt to circumvent the platform's governance, refusal, authority or rate controls, including by prompt manipulation.
  • Scrape, reverse engineer, or attempt to extract prompts, weights, proprietary mechanisms or other users' data.
  • Submit malicious content, or use the platform to attack, degrade or overload any system.

Breach of this section is a material breach of the Terms & Conditions and may lead to suspension or termination.

06

Human oversight and your right to review

Under section 12.1 of Quebec Law 25, where a decision about an individual is based exclusively on automated processing and has a significant effect on them, that person must be informed, told the personal information used and the principal factors that led to the decision, permitted to submit observations, and given review by a person with authority to correct the decision. Law 25 reaches any organisation holding the personal information of Quebec residents, so it applies to us as a British Columbia company.

Whether ATORA makes any decision exclusively by automated means about an individual: [ AUTOMATED DECISION SCOPE ].

Regardless of that answer, our design keeps a human in the loop for material actions, and a language model never determines a probability, forecast or confidence figure.

To request human review of a result that affects you, write to hello@atora.ai. We will tell you what personal information was used and the principal factors that produced the result, accept your observations, and have a person with authority to correct the outcome consider them. You will get a substantive response within 30 days.

07

Reporting a concern

  • Misuse of the platform, or output you believe is harmful — hello@atora.ai.
  • Privacy concern, or a request for human review — hello@atora.ai, addressed to [ PRIVACY OFFICER ].
  • Security vulnerability — hello@atora.ai. See our Security page for the disclosure process.

We acknowledge reports within [ RESPONSE WINDOW ]. A report that reveals a real defect becomes a recorded defect with an owner, and we will tell you what changed.

Questions about this page? hello@atora.ai