AI Usage Disclosure

1. Short version

We use AI tools to assist with analysis and drafting. We do not permit client data to be used to train models. We do not put client confidential information into any AI tool whose terms fail to prohibit training on that input. A person reviews everything before it reaches a client.

2. The training prohibition

This is the constraint that governs everything else on this page, and it is not negotiable per engagement:

This applies to models we operate, models operated by vendors, and any future model. It applies to the raw material a client gives us and to derivatives of it. It survives the end of the engagement.

3. How we select tools

Before client confidential information touches a tool, we check that its terms contractually prohibit training on submitted input, that data is not retained beyond what is operationally necessary, that any human review by the vendor is disclosed and bounded, and that the commitment is in the enforceable terms rather than only in marketing material.

Where a tool cannot meet that bar, we either do not use it, or use it only with information that is not confidential. Free and consumer tiers of consumer AI products frequently reserve training rights, and are not used with client material.

4. Where AI is used

Typically: structuring and drafting sections of the audit report; summarising and organising notes we ourselves produced; researching publicly available technical documentation; generating and reviewing example code, configurations, and architecture options; and checking our own work for gaps and internal inconsistency.

In all of these, AI accelerates work a person directs, reviews, and remains responsible for.

5. Where AI is not used

AI does not make the go/no-go recommendation. AI does not decide what is in scope. AI does not set the payback figures — those come from client-supplied inputs applied by a person. AI-generated content is never delivered to a client unreviewed, and no deliverable is machine-generated end to end.

We do not use AI to evaluate, score, or make decisions about individual people, and the audit produces no such assessment.

6. Human review

Every deliverable is reviewed by the practitioner running the engagement before delivery. Technical claims are checked against source documentation or tested directly rather than accepted because a model asserted them. Figures are checked against the client's own inputs. Where we cannot verify something, the report says so rather than smoothing it over.

7. Known limitations

AI systems produce confident, fluent output that can be wrong, can fabricate citations and technical details, and can reflect biases in training data. Human review reduces this risk substantially. It does not eliminate it, and we will not claim that it does.

If you find an error in a deliverable, tell us and we will correct it and reissue the document.

8. What we ask of clients

Send us only what the audit needs. Redact client names, matter identifiers, and personal information from sample documents where the substance survives redaction — we are usually assessing structure and process, not content. Tell us if any material carries obligations beyond ordinary confidentiality, such as privilege, a protective order, or a third-party contractual restriction, so we can handle it accordingly or decline to receive it.

Do not send protected health information. Healthcare engagements are not currently being accepted.

9. Changes

Tooling changes as the market changes. The prohibition in section 2 does not. If our tooling changes materially during your engagement, we will tell you.

10. Contact

Questions about this disclosure, or about the specific tools in use for your engagement: contact@beyondxai.com. BeyondX LLC, 1930 Burbank Ave NW, Olympia, WA 98502, United States. +1 360 909 9015 (voice and WhatsApp).