Per OpenAI's announcement, GPT-5.6 is now the preferred model powering Microsoft 365 Copilot across Word, Excel, PowerPoint, Chat, and Cowork. For Australian businesses that live inside the Microsoft stack, it can feel like the decision has been made for you. It has not. The model behind the Copilot button is one workload. The question of which frontier model does your highest-stakes work is still open, and it is worth answering deliberately rather than inheriting whatever ships by default.
What actually changed
The headline is a default-model swap inside Microsoft 365 Copilot. The assistant surface embedded in the Office apps most Australian teams already pay for now leads with GPT-5.6. In practice that means the AI behind 'summarise this document' and 'draft this email' has changed underneath, often without anyone in the business choosing it. That is a distribution event: OpenAI reaching millions of desks through Microsoft's install base. It is not, on its own, a statement about which model is most reliable for the work that actually matters to you. A bigger reach is not the same thing as a better result on your hardest tasks. For a business, the practical point is simple: the button changed, but your evaluation of what to trust with real work should not change with it.
Why a default is not a decision
Default and decided are two different things. Claude is already generally available in Microsoft Foundry, which means Microsoft-stack businesses can build on Claude inside the same cloud and identity perimeter they already trust. So the real question is not what sits behind the Copilot button. It is which model you want running your reasoning-heavy, governance-sensitive work, and those can be different answers. Three things are worth weighing before you accept the default:
The Copilot surface is one workload, not all of them. In-app summarise and draft is genuinely useful and low-stakes. The work that moves the needle, multi-step agents, document-heavy workflows, code, and customer-facing automation, is where model choice matters, and where you are free to run Claude through Foundry or directly.
Reliability and data posture beat whatever is built in. For Australian businesses the questions that count are whether the model holds a long task together without quiet failures, and where your data sits under the Privacy Act and who can see it. Those get evaluated per workload, not settled by a default.
Single-vendor lock-in by default is still a choice. Standardising your whole AI strategy on whatever your productivity suite happens to ship is a decision made for you. A deliberate mix, often with Claude on the complex and sensitive work, keeps control on your side.
How to choose per workload, not per app
The practical move for an Australian team is to stop treating 'our AI' as one switch and start treating it as a set of decisions tied to specific jobs. Keep Copilot for the light in-app tasks it does well. For the workloads where a mistake is expensive or the data is sensitive, run a short evaluation and pick the model that proves itself, rather than accepting the one that arrived with an update. A simple sequence works:
List your top five AI workloads by stakes, not by frequency. The daily email draft is frequent and cheap to get wrong. The quarterly board report or the customer refund workflow is not.
For each high-stakes workload, test Claude and the Copilot default on the same real task, and compare reliability rather than demo polish.
Check where the data goes and whether it stays inside your existing Microsoft identity and compliance boundary. Claude in Microsoft Foundry keeps that boundary intact.
Write the decision down, so the next person does not silently inherit a default again.
What indecision quietly costs
Doing nothing is not free. Say a 30-person Sydney firm lets the default model handle a document-heavy workflow it is not reliable enough for, and each week that produces two hours of rework across the team. At a blended rate of $150 an hour, that is roughly $15,600 a year quietly lost to a decision nobody actually made. The fix is not expensive. It is an afternoon of structured testing and one written call about which model runs which job. Spend the afternoon once, and the saving repeats every week after that.
OpenAI becoming the Copilot default is a good prompt to make a real decision instead of inheriting one. Because Claude is available in Microsoft Foundry, Australian businesses on the Microsoft stack can keep Copilot for the light in-app tasks while running Claude where reliability, data governance, and complex automation matter. Automata AI helps Australian teams make that call and set it up. The goal is a stack you chose on purpose, workload by workload, rather than one assembled for you by whichever vendor shipped last.Book a brainstorm.
