← All episodes
Special · News ·9:03 ·June 3, 2026

Microsoft Is Replacing OpenAI Inside Copilot — and You Don't Get a Vote

In August, Microsoft swaps the OpenAI model inside GitHub Copilot for its own and migrates 26 million developers by default. Treat it as a settings update and you'll miss the supply-chain substitution underneath.

The Promise

PROMISE RISK
Leans Risk

The Risk

A default swap is a supply-chain event

In August, Microsoft replaces the OpenAI model inside GitHub Copilot with Project Polaris — its own coding model, MAI-Code-1. Twenty-six million developers get migrated by default. Almost none of them get a say.

The instinct is to file this under settings. A model changed; the autocomplete still works; move on. That instinct is the problem. The component that writes a measurable share of your codebase just got swapped for a different supplier, on a different training set, with different failure modes. In any other part of the business, replacing a supplier that touches production gets logged as a change. This one won’t, unless someone decides it should.

”You can switch back” is the false comfort

Microsoft is offering a three-month fallback window. The reassurance writes itself: nothing is forced, you can switch back. After 25 years in cybersecurity I have learned to read a fallback window for what it is — a countdown, not a guarantee. Defaults win. Three months from now the OpenAI option is a setting most teams have forgotten exists, and the muscle memory, the prompt patterns, and the review habits have all reformed around Polaris. The exit you were promised quietly closes on its own.

One company, three layers

Project Polaris is the visible part of a larger move: Microsoft decoupling from OpenAI and building its own stack. Follow it down and you reach a single vendor that owns the model, the chips it runs on, and the cloud it runs in. That is the kind of concentration that looks like convenience right up until the day it looks like exposure. One roadmap, one outage domain, one pricing lever, sitting across the tool your developers touch every hour.

Three questions before August

One. If our code assistant changes models, does anyone in this organization log it as a change? Two. What did we actually validate about the new model, versus assuming parity because the vendor name stayed the same? Three. Where else have we accepted a default that quietly made one company our model, our compute, and our cloud all at once? Ask them in the next operating review. The swap is coming whether or not the question gets asked first.