The RCMG Responsible AI Standard

1. 01

   Use AI intentionally.

   We don’t adopt AI simply because a tool exists or a task can be automated.

   There should be a reason.

   AI should improve the quality of the work, expand human capability, increase appropriate efficiency, solve a meaningful problem, or create a better experience.

   The question isn’t only, Can AI do this?

   We also ask, Should it? And under what conditions?

2. 02

   Protect people and information.

   Every piece of information doesn’t belong in an AI system.

   We consider privacy, confidentiality, data sensitivity, system permissions, vendor practices, and client requirements before information is shared with or made accessible to AI.

   As the sensitivity of the information increases, so should the safeguards around it.

3. 03

   Verify what matters.

   AI can produce an answer that sounds certain and still be wrong. We don’t treat confidence as accuracy.

   Important facts, statistics, research, citations, recommendations, and other consequential information receive an appropriate level of verification before they are relied upon.

   The greater the consequence of being wrong, the greater the level of review.

4. 04

   Maintain meaningful human judgment.

   AI can assist with analysis, recommendations, pattern recognition, research, and decision support.

   It doesn’t remove human accountability.

   Substantive client work receives human review, and we don’t allow AI to independently make consequential decisions where meaningful human judgment is required.

   Human oversight should be real—not a checkbox added after the technology makes the decision.

5. 05

   Make claims we can support.

   We work in a market full of big promises about AI.

   We don’t add to them.

   When we describe an AI tool, workflow, product, or service, we aim to accurately represent what it can do, what it can’t do, how much human involvement remains, and what someone should reasonably expect from it.

   A model being technically capable of doing something doesn’t automatically mean a particular implementation does it reliably.

6. 06

   Don’t confuse prediction with understanding people.

   AI can identify patterns, generate possibilities, and help develop hypotheses about customers and audiences.

   But a generated persona isn’t a customer; a prediction isn’t proof.

   And an AI-generated insight isn’t automatically evidence about real people.

   When we use AI for customer, audience, or behavioral analysis, those outputs should be strengthened with actual research, data, feedback, observation, and human understanding when appropriate.

7. 07

   Be transparent when transparency matters.

   We don’t believe responsible AI means announcing every time technology helps edit a sentence, brainstorm an idea, or organize information.

   Transparency should be meaningful.

   When AI plays a material role in an interaction, recommendation, decision, or synthetic experience where its involvement would reasonably matter, people should understand that AI is involved.

   If you’re interacting directly with an automated AI experience we provide, we don’t want you wondering if you’re talking to a human.

8. 08

   Respect content provenance.

   AI platforms are using machine-readable markings, content credentials, watermarks, and other provenance mechanisms to communicate when AI has generated or processed content.

   Where those mechanisms are present, we don’t intentionally defeat them for the purpose of hiding material AI involvement or misleading an audience.

   We also don’t assume AI-detection technology is infallible.

   A technical signal can provide useful information without telling the entire story of who created a work, where the ideas originated, or how much human contribution occurred.

9. 09

   Give AI only the authority it needs.

   There’s a meaningful difference between an AI system that generates information and one that can take action.

   AI agents may be able to send messages, access files, update records, interact with software, publish information, or execute multi-step workflows.

   That additional capability requires additional governance.

   We apply the principle of minimum necessary authority: an AI system should have the information, permissions, and autonomy it needs to perform its approved function—and no more.

   Higher-risk actions require stronger controls, traceability, and human oversight.

10. 10

    Build responsibly, too.

    There’s a difference between using an AI tool and putting an AI-powered system into someone else’s hands.

    RCMG develops and configures AI-enabled assistants, assessments, workflows, applications, and other business tools.

    When we build them, we consider more than whether the technology works.

    We consider:

    • —What information the system can access
    • —What actions it can take
    • —Who will use it
    • —Where human judgment belongs
    • —What happens when it’s wrong
    • —How users know AI is involved
    • —How data should be protected
    • —What permissions are necessary
    • —What limitations users need to understand
    • —What responsibilities belong to RCMG, our technology providers, and our clients

    Responsible AI doesn’t stop at the model, it includes the system built around it.

11. 11

    Increase oversight as risk increases.

    Not every use of AI deserves the same level of governance.

    Brainstorming a headline and making an employment decision aren’t equivalent.

    Editing a paragraph and allowing an autonomous agent to modify business records aren’t equivalent.

    Our approach is risk-based.

    As sensitivity, autonomy, access, potential harm, or consequences increase, so do our requirements for verification, security, transparency, and human oversight.

• Use AI intentionally.
• Protect people and information.
• Verify what matters.
• Make claims you can support.
• Maintain meaningful human judgment.
• Be transparent when transparency matters.
• Respect legitimate signals about content provenance.
• Give AI only the authority it needs.
• Don’t confuse automation with accountability.

And above all:

AI can support the work. Humans own the result.

The RCMG Responsible AI Standard is informed by our internal Responsible AI Use & Governance Policy and evolves as AI technology, accepted practices, and the risks surrounding its use evolve.

Last updated: October 2026