> ## Documentation Index
> Fetch the complete documentation index at: https://docs.aperium.apps.hillspire.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Responsible & safe use

> What data is OK to use, and how to verify outputs before trusting them.

This page is about one habit: a quick gut check you run before you paste anything into an AI tool, and before you trust anything that comes back. It takes a few seconds and it prevents the moment where a client's portfolio holdings end up on a third party's servers because pasting them felt like using a search engine.

## Know which tool you're in

The single biggest lever is knowing whether you're in Aperium or a public tool.

<CardGroup cols={2}>
  <Card title="Aperium — internal" icon="building-shield">
    Hillspire's own platform: hosted for us, connected only to our approved
    internal systems, inside our security boundary. Closer to an internal
    channel than a public forum. Safer for internal data.
  </Card>

  <Card title="Public tools" icon="globe">
    The free ChatGPT app, Gemini, and anything you'd open directly on the open
    web. A third party's product, under a consumer terms-of-service, with no
    guarantee about how your input is retained or used. Treat it like posting
    on the open internet.
  </Card>
</CardGroup>

The rule isn't "never put in company data." It's "know which tool you're in, because the two are not the same risk."

## Generally fine to enter — in either tool

<Check>
  * **Public information** — already-published filings, press releases, marketing copy.
  * **Your own drafts and ideas** — an email or memo you're writing.
  * **General questions** — "how do I structure this," "explain this concept" — with no firm or client specifics.
</Check>

Most day-to-day AI use lives here and needs no second thought.

## Needs care — even inside Aperium

Four categories deserve a pause, framed for a financial and investment firm:

| Category                    | What it covers                                                                                                                                                             |
| --------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **MNPI**                    | Material non-public information — anything that could move a security's price if it leaked: pending deal terms, unannounced earnings, non-public portfolio-company results |
| **PII**                     | Personally identifiable information — SSNs, dates of birth, account numbers, personal financial details of employees or clients                                            |
| **Client & portfolio data** | Specific holdings, NAV, performance figures, LP / investor details                                                                                                         |
| **NDA material**            | Anything under a signed confidentiality agreement with a counterparty                                                                                                      |

<Warning>
  Some of this may be fine to reference **inside Aperium** — it's our internal,
  connected platform — but never in a public tool. When in doubt, treat it as
  sensitive and ask before entering it.
</Warning>

## The decision checklist

Run these three questions, in order, before pasting anything:

<Steps>
  <Step title="Am I in Aperium or a public tool?">
    A public tool means public information only — full stop.
  </Step>

  <Step title="Does this touch MNPI, PII, client data, or NDA material?">
    If yes, pause and confirm before entering it — even inside Aperium.
  </Step>

  <Step title="Am I about to trust or forward the output without checking it?">
    If so, verify first (see below).
  </Step>
</Steps>

<Info>
  Unsure on any of the three? Ask before you paste.
</Info>

### Two worked scenarios

<AccordionGroup>
  <Accordion title="Quick client summary">
    Prepping for a call, you're tempted to paste a client's account data into
    the public ChatGPT app to save time. The checklist: **public tool + client
    and portfolio data = stop.** Use an approved internal path instead, or
    describe the task generically — "help me structure a client performance
    summary" works fine without the real figures in the prompt.
  </Accordion>

  <Accordion title="Early word on a deal">
    You've heard early, non-public details of a pending acquisition (MNPI) and
    want help drafting talking points — and you have Aperium open. The
    checklist: **internal tool, but still MNPI = pause and confirm** before
    entering deal specifics, even here. Being on the approved platform doesn't
    erase the need for care on the most sensitive category.
  </Accordion>
</AccordionGroup>

## Verify before you trust

An AI model produces plausible text, not verified fact — confident does not mean correct. Before you trust or forward any output that contains a figure, a quote, a citation, or a claim you didn't personally check, verify it against the source.

<Tip>
  Treat any output like a first draft from a very capable colleague who is
  occasionally confidently wrong: useful, but not authoritative until you've
  checked it. Turn the scrutiny up further for anything going external, feeding
  a client-facing number, or informing a real decision.
</Tip>

## Privacy & compliance basics

A few principles hold regardless of the exact policy text:

* Use the least data necessary to get the answer you need.
* Match the tool to the data's sensitivity.
* Ask before you paste if you're not sure.

<Info>
  Confirm the specifics against Hillspire's official AI-use and data-handling
  policy — the approved-tools list, the precise do's and don'ts by data
  category, and who to contact with questions.
</Info>