Watney proposes. A person decides.
Watney is the AI assistant inside Backsplice. It drafts model and system cards, suggests how to triage a new AI system, answers questions from your organization's own records, reviews assessment answers against their evidence, and suggests what to do next. Every output is a proposal. It does not approve a system, set a risk tier or publish a card; those are decisions a person makes and the record shows who made them.
Read Watney's own system and model cards: Watney is an AI system, and we govern it the way we help you govern yours. The cards name the model provider, the models in use and the provider's terms.
Approving an AI system is a judgment. Software should not pretend otherwise.
When your organization approves an AI system, it is stating that it looked at the system, weighed the risk and accepted it. That judgment belongs to named people, with a reason on the record. "The software decided" is not a reason anyone can defend.
Watney takes on the slow parts of the work: the first draft of a card, a suggested risk tier for a new system, a read of whether the evidence supports an answer. Then it stops, because the next step is a decision and decisions belong to your team.
Four controls, enforced in the product
These are not policy statements. They are how the software behaves.
An admin opts in first
Watney does not read your organization's data until an org admin opts in, on a settings screen that states what would be sent to the model provider. An admin can switch it off again at any time.
It does not overwrite your answer
When Watney reviews an assessment answer, its verdict and its reasoning are stored beside your answer rather than on top of it. Your answer stays the one you gave. The review is an annotation you are free to disagree with.
You can overrule it, on the record
Disagree with a verdict and you overrule it with a note saying why. The override is stored with your name and the time, so the disagreement becomes part of the record rather than a silent edit.
A new review needs new attention
When Watney reviews an answer again, its earlier verdict is replaced and your previous override is cleared. A judgment you made about the old evidence is not allowed to stand as approval of the new.
Drafts work the same way. Watney can write a model card, a system card, a risk entry or a suggested answer, and each one lands in front of you as a proposal. If nobody saves it, nothing was written.
The rules are code, on purpose
Where a question has a right answer that can be computed, Backsplice computes it. Watney helps where the work is reading and drafting.
Answered by code
Deterministic, repeatable, identical every time you ask.
- Who has to approve. Your risk-tier policy says whether any, a majority or all of the approvers must agree. The count is arithmetic.
- When a re-review is due. Each approval carries a review date set by its risk tier; when it arrives a re-review opens automatically, and the system keeps its stage until that review decides. That is a date comparison.
- Whether a published card changed. Publishing freezes a version with a tamper-evident seal, checked by code.
- Whether the audit log was altered. Each entry is hash-chained to the one before it, so a later change to a recorded entry is detectable unless it is made by someone holding the application's signing key and database access. That makes the log tamper-evident. It is not immutable.
Helped by Watney
Where the work needs reading comprehension, or a first draft.
- Write me a first draft. Model and system cards you then edit into something true for your system.
- How should we triage this system? A suggested risk tier for a new system at intake. A person sets the tier.
- Does this evidence support this answer? Comparing a document against a question is reading work, and it is the slowest part of an assessment.
- What does our program say about this? Answers grounded on your organization's own records.
- What should we do next? Suggested next actions drawn from your open work.
The dividing line is simple. If a question has a correct answer that can be computed, we compute it. If it needs someone to read a document and form a view, Watney drafts the view and a person confirms it.
What Watney is not allowed to do
Restrictions built into the product, not intentions we hold.
- Approving a system. Approve, approve with conditions, reject and send back are recorded only inside a review, by an approver, with a reason. Watney has no path to a decision.
- Setting a risk tier. It can suggest one at intake. The tier on the record is the one a person set.
- Publishing a card. It can draft a model or system card. Publishing, which freezes and seals a version, is a person's act.
- Taking instructions from your documents. Everything you upload or paste is treated as material to analyze rather than as commands. Text inside a file that tries to steer a verdict is treated as a warning sign and resolved conservatively.
- Filling gaps with plausible guesses. Watney is instructed to base what it says on your records and, where they do not support a conclusion, to say that the information is missing.
Watney runs only where it has been switched on, and every use is recorded
Watney comes with a plan, and an org admin decides whether your organization uses it.
An admin opts in
An org admin gives consent before Watney reads your organization's data, and can withdraw it. Everything else in Backsplice works without Watney: approvals, registers and exports never depend on it.
Included on both plans
Watney comes with Premium and Enterprise at no extra charge, within fair-use limits. If a limit is reached, Watney pauses, and the pause message says when it resumes or how to turn it back on; everything else keeps working. Compare the plans.
Not in the demo
Watney, exports and scheduled report emails, auditor and guest links, the transparency page and public badge, the API and MCP server, webhooks and SIEM forwarding come with a plan.
Backsplice's own staff use a separate staff assistant. Your organization's opt-in does not apply to it; Watney's system card describes it.
What is sent, and to whom
Watney sends the model provider only what a task needs: the records your team typed, documentation you paste in, and the contents of uploaded evidence files (or its earlier summaries of them) when Watney reviews or drafts an answer or summarizes evidence. Records can name people, so anything typed into a record can reach the provider.
The model provider, the models in use and the provider's own terms for that data are stated on Watney's system card. Read them before you opt in. If they do not fit your organization, leave Watney off.
Judge Watney on its drafts
If you would rather go through the controls with a person first, ask us and we will walk you through them.
Watney is included on Premium and Enterprise. It is not part of the demo. Compare the plans.