> For the complete documentation index, see [llms.txt](https://handbook.sixpivot.com.au/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://handbook.sixpivot.com.au/~/revisions/Srsqj7Icjx72OqJmSZiY/how-we-work/policies/responsible-use-of-ai.md). # Responsible Use of AI How we use AI responsibly at SixPivot. ## Why this exists AI is part of how we work at SixPivot. We encourage Pivots to explore and use AI where it helps us work more effectively, whether that's writing software, researching a problem, analysing information, creating content or automating parts of our work. With that capability comes responsibility. Using AI can mean sharing information with another service, giving a tool access to a codebase or allowing an agent to take actions on our behalf. We need to understand those boundaries and make sensible decisions about what we give AI access to. The principle is simple: > **Know what information you're giving AI, where that information is going and what the AI is allowed to do.** AI can help produce the work. It doesn't change who is responsible for it. ## Our approach to AI We don't want people to avoid AI because there is a policy around it. Quite the opposite. We expect Pivots to understand and make appropriate use of the technology available to them. Experimentation and learning are an important part of keeping our skills current and understanding where AI genuinely helps. At the same time, using AI doesn't lower our expectations around quality, security, privacy or professional judgement. When using AI for SixPivot or client work: * understand the tools and services you're using; * consider what information you're providing them; * respect client policies and confidentiality; * use an appropriate account and environment for the work; * review and take responsibility for AI-assisted work; and * ask when you're unsure. Our existing [Cybersecurity Policy](https://handbook.sixpivot.com.au/how-we-work/policies/cybersecurity-policy) continues to apply when using AI. ## Think beyond the AI tool It's easy to think about AI in terms of the application in front of you: ChatGPT, Claude, Cursor, Copilot or whatever comes next. The actual path your information takes can be more complicated and nuanced. When evaluating AI use, consider three parts: ### Model The **model** is the AI model ultimately processing your request - Eg. Fable, Opus, Luna, Qwen. Consider: * what model you're using; * whether it's appropriate for the task; * whether there are licensing or usage restrictions relevant to your work; and * whether the model is running locally or remotely. The organisation that created a model isn't necessarily the organisation processing your data. ### Provider The **provider** is the service actually running/hosting the model and processing your request. Consider: * who is processing the information; * where the information is processed and stored; * whether it may cross regions or jurisdictions; * what information is retained and for how long; * whether inputs or outputs may be used for training; and * what contractual, privacy and security protections apply. The same model can be available through several providers with very different privacy, security and data sovereignty arrangements. Eg. Anthropic (US) vs Azure Foundry (AU) ### Tool The **tool** is how you interact with the model - Eg. ChatGPT, Claude Code, Cursor or Junie. Consider: * what information the tool sends to the model or provider; * whether it automatically includes additional context; * whether it indexes or uploads files; * what parts of your filesystem or environment it can access; * whether it can execute commands; * what plugins, MCP servers or connectors it can access; and * whether it can take actions on your behalf. An approved tool doesn't mean every feature, provider or integration available through that tool is appropriate for every situation. ### Use the right model for the work Bigger or more capable models aren't automatically the right choice for every task. Consider the capability required, the amount of context being processed, how often the task will run and the cost of using the model. Routine tasks may be better suited to smaller or faster models, while complex reasoning or particularly difficult engineering work may justify a more capable model. This becomes especially important when building automated or agentic workflows where a decision made once can result in thousands of model calls. We don't expect Pivots to optimise every prompt for cost. We do expect people to make sensible choices and avoid unnecessarily expensive use where a cheaper model or approach produces an equivalent outcome. Eg. Using Fable to change the colour of a border is in-efficient, using Sonnet would work optimally. ## Think about the data Across all three layers, consider the information you're working with. Not all information carries the same risk. Be particularly careful with: * client confidential information; * source code and intellectual property; * personal or sensitive information; * production data; * credentials, secrets and API keys; * commercially sensitive SixPivot information; and * information subject to regulatory or data sovereignty requirements. Only provide AI systems with the information reasonably required for the task. If you wouldn't be comfortable sharing the information with the provider under the applicable client and SixPivot arrangements, don't put it into the tool. When in doubt, as on our #ai Slack Channel or reach out to the PLT. ### Before using AI For everyday use this should become a quick mental check rather than a lengthy approval process. #### Data & Privacy * [ ] Am I allowed to provide this information to this AI service? * [ ] Does it contain client confidential, personal, sensitive or otherwise restricted information? * [ ] Does the client have specific requirements for AI use? * [ ] Am I sharing only what is reasonably needed for the task? #### Cost & Effectiveness * [ ] Am I using an appropriate model for the complexity of the task? * [ ] Am I providing more context or making more model calls than reasonably needed? * [ ] If this is automated or agentic, are loops, retries and parallel work appropriately bounded? * [ ] Is the expected cost reasonable for the value of the task? #### Security * [ ] Do I understand what the AI can access and what actions it can take? * [ ] Are its permissions, integrations and connectors appropriate for this task? * [ ] Could it access credentials, secrets or systems that it doesn't need? * [ ] Are sensitive, consequential or destructive actions appropriately restricted? * [ ] Am I using MCPs, Skills or Agents that may transmit or call services that may have unknown or restricted behaviours? If you can't confidently answer these questions, find out before providing client or sensitive information to the service. > **An approved model doesn't make every provider appropriate. An approved provider doesn't make every tool or integration appropriate. An approved tool doesn't mean every type of data can be put into it.** ## Client engagements Our clients have different approaches to AI. Some actively encourage its use. Others restrict particular models, providers or tools. Some prohibit AI from processing certain classes of information entirely. When working with a client, their requirements for their systems and information take precedence. Don't assume that because a tool is available or approved for SixPivot use that it's automatically appropriate for client work. We also have to appreciate and understand [Device Policy](/~/revisions/Srsqj7Icjx72OqJmSZiY/how-we-work/policies/device-security-and-client-devices.md) and [Client Devices](/~/revisions/Srsqj7Icjx72OqJmSZiY/how-we-work/policies/device-security-and-client-devices.md#when-a-client-gives-you-a-device) and their operating environments too. ### Before an engagement Understand the client's expectations around AI before using it with their information. Where relevant, establish: * whether the client has an AI or acceptable-use policy; * which AI tools and services are permitted; * whether particular models or providers are restricted; * what types of client information can be processed; * whether there are data sovereignty or geographic processing requirements; * whether you should use SixPivot tooling or a client-provided environment; and * who to ask when the requirements aren't clear. This is particularly important where AI is going to be a significant part of how the engagement is delivered. ### During an engagement Stay within the boundaries established for the engagement. Use the appropriate accounts, tools and providers for the work and avoid unnecessarily moving client information between AI services. Be aware that the configuration of AI tools can change. Adding a new provider, enabling a connector or giving an agent additional permissions can materially change where information goes or what the tool can access. If the way you're using AI changes significantly during an engagement, reconsider the Model, Provider, Tool and Data questions rather than assuming the original decision still applies. ### When an engagement ends AI environments can retain more than conversation history. Where appropriate, clean up client-specific: * projects and workspaces; * uploaded files and documents; * conversation history or persistent context; * codebase indexes or embeddings; * local models or data stores containing client information; * MCP servers and connectors; * credentials and API keys; and * agent configuration created for the engagement. There may be legitimate reasons to retain information for ongoing support or other agreed purposes. Follow the arrangements for the engagement and our other policies where this applies. ## You are responsible for the outcome AI-generated work is still our work. Whether AI helped write code, analyse information, draft a document or make a recommendation, you remain responsible for deciding whether the result is appropriate to use. That means applying the same professional judgement we would to work produced without AI. AI output can be incorrect, incomplete or misleading while appearing convincing. Review it to a level appropriate for the consequence of getting it wrong. ### AI with access and agency There is an important difference between an AI that can answer a question and one that can interact with your environment. Coding agents, MCP servers, plugins, connectors, skills and other integrations can give AI access to files, source repositories, shells, databases, browsers, cloud environments and other systems. The more capability an AI system has, the more carefully its access should be considered. Don't give an AI system access simply because the integration is convenient. Understand what it can see, what it can change and which trust boundaries that access crosses. Our Agents and Connected AI guidance covers this in more detail. ### If something goes wrong We're human, mistakes happen. If you accidentally provide information to an inappropriate AI service, expose credentials, discover that information has been processed somewhere unexpected or believe an AI tool may have caused a security or privacy issue, raise it promptly. Follow our [Cybersecurity Policy](https://handbook.sixpivot.com.au/how-we-work/policies/cybersecurity-policy) and any applicable client incident process. Don't try to quietly fix or hide an incident. Early reporting gives SixPivot and the client the best opportunity to understand what happened and manage any risk. ### In short **Understand the path** Think about the Model, Provider and Tool, not just the AI application in front of you. **Protect the data** Know what you're sharing, minimise unnecessary information and respect client requirements. **Understand the capability** Know what the AI can access and what it can do on your behalf. **Understand the security** Know the security implications for the use of AI **Own the outcome** AI can assist with our work, but responsibility for the work remains with us. **When in doubt, ask** If you're unsure whether a particular model, provider, tool or use of client information is appropriate, find out before using it. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://handbook.sixpivot.com.au/~/revisions/Srsqj7Icjx72OqJmSZiY/how-we-work/policies/responsible-use-of-ai.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.