Artificial Intelligence Use Policy

Interactive Guru, LLC
Adopted September 15, 2026 · Revised September 28, 2026
IG-POL-AI-01

Interactive Guru, LLC is an Oregon domestic limited liability company. This policy applies to Christopher A. Brown and to every contractor, subcontractor and partner studio performing work on an Interactive Guru engagement.

1. What this policy is

Interactive Guru is a one-person studio that works with contractors and partner studios. This policy states the rules that already govern how AI tools are used here. It is short because the studio is small and because every rule in it is one we actually follow. It does not describe a review board, a training curriculum, an audit programme or a certification, because we have none of those things.

One person is accountable for every deliverable that leaves this studio: Christopher A. Brown. That accountability does not change, transfer or dilute because a tool was used to produce part of the work. There is no such thing here as an output that "the AI produced."

2. The four rules

2.1 A human authors, reviews and approves everything before it reaches a client. No AI output is delivered as found. Generative output is treated as raw material, a draft, a concept or a starting frame, that a human then edits, art-directs, rewrites or discards. No deliverable ships on an AI tool's say-so, and no AI system makes a final creative decision, approves a deliverable, or signs off on quality. Section 3 states exactly which stage of our process is that human approval, and which stages are not.

2.2 Client and confidential information does not go into unapproved tools. Confidential client material, personal information, credentials and regulated data are not submitted to any third-party AI service that the client has not approved. Where that kind of work requires AI assistance, it is done on local, on-device models that transmit nothing off our own hardware, or with AI assistance switched off entirely. We do not use client data or client deliverables to train any model, and we do not accept vendor terms that permit a vendor to do so.

2.3 Facts are traced to a source, never generated. Where a claim about a client, a market, a regulation or a technical standard appears in our work, it is carried from an identified source document, not produced by a language model and left to stand. If a fact cannot be evidenced, the claim is removed rather than softened or hedged. We do not assert facts on a client's behalf that only the client holds.

2.4 Material AI use is disclosed, not concealed. Where AI materially contributed to a deliverable, we tell the client, per deliverable, and we keep our own record of which assets in a deliverable were AI-assisted. We do not present AI-assisted work as though it were not, and we do not present it as a selling point either. It is a fact about how the work was made, and clients are entitled to it.

3. How the work is actually checked

Interactive Guru runs a written, staged verification process on production work, and the stages are separated on purpose so that nobody grades their own homework:

1. The producer verifies their own work against a written specification and records the evidence, the rendered page, the screenshot, the contract check, the test run, rather than asserting that it is done.
2. A separate, independent review pass reads the specification, the artifact and the named evidence. It reads the work, not the producer's description of the work, and it returns a pass or a specific list of required changes. It cannot be skipped because a job looks small.
3. Christopher A. Brown then opens the actual deliverable, renders the PDF, views the image, reads the copy, and accepts or rejects it against the brief and the client's intent. A verification report is never a substitute for looking at the thing.

Stages 1 and 2 are performed by AI agents working to written procedures under Christopher A. Brown's direction, each starting from the specification, the artifact and the recorded evidence rather than from the previous stage's account of it.

They are internal checks. Neither of them is the approval that rule 2.1 requires: that approval is stage 3, and it is a person, Christopher A. Brown, reading or viewing the finished deliverable itself. Nothing reaches a client on the strength of stages 1 and 2 alone. The point of the sequence is that AI-assisted output is checked by a process that assumes it may be wrong, and that a named human is accountable for each deliverable and has looked at the finished work before a client does.

4. Copyright and provenance

1. We do not prompt a generative tool with a living artist's or a competitor's name in order to imitate their style.
2. We do not use generative tools to reproduce or approximate an identifiable existing work, character, logo or protected mark.
3. Generative output is art-directed and reworked, not delivered as found.
4. Image assets are reverse-image checked before they go into a client deliverable.
5. Where provenance matters, we use licensed stock or original work instead of generation.
6. We use tools whose terms assign ownership of the output to the customer.
7. We keep the record of which assets in a deliverable were AI-assisted, so that ownership and licensing can be stated accurately in a delivery.

5. Tools in use, by stage

This is what we use today. It changes as tools change; the rules in section 2 do not.

6. Client-specific rules take precedence

Where a client imposes stricter requirements, an approved-tool list, a prohibition, a review or approval step, a security review, or contract terms governing AI, those requirements govern and this policy yields to them. Where a client requires review or approval before an AI tool is used on their data, systems, networks or deliverables, that approval is obtained before the use begins, not disclosed afterward. For any client who asks, we are prepared to provide, per tool: the tool and model, the vendor, the use case, the data types involved, hosting, retention and model-training terms, the security controls, the human oversight applied, the intellectual-property treatment of the output, how the output is validated, and any additional safeguards proposed.

7. Review and document control

This policy is reviewed by Christopher A. Brown at least annually, and whenever a tool, a vendor term, or a client requirement changes in a way that affects it.