Open your favorite code editor and check who actually made the AI inside it. Not the company logo on the toolbar — the model doing the work. You might be surprised.

Six months ago, the answer was simple. Copilot ran OpenAI. Cursor ran OpenAI. Claude Code ran Claude. Brand equaled model. You picked a tool, and you knew whose AI you were trusting with your codebase.

April 2026: The Supply Chain Blew Open

Cursor 3 had already set the pattern. On April 2, it launched Composer 2, a code-generating agent built on Kimi K2.5 — a base model from Moonshot AI, a Chinese startup most Western developers have never heard of. Cursor took that model and fine-tuned it with four rounds of RL (reinforcement learning — the training method where AI learns by trial and error rather than memorizing examples). The result works well. The provenance raises questions nobody's asking yet.

Two weeks later, the rest of the industry caught up.

On April 14, GitHub enabled a model picker inside Copilot. Users can now choose Claude — Sonnet 4.5, 4.6, Opus 4.5, 4.6 — right alongside OpenAI's own Codex agents. This is the tool Microsoft built on a $13 billion OpenAI investment, now serving its investor's primary competitor with equal billing. Two days later, on April 16, GitHub rolled out Claude Opus 4.7 across Pro+, Business, and Enterprise tiers. Anthropic's newest model, inside Microsoft's product, available to paying customers on launch day.

On April 15, OpenAI shipped something even stranger. Their Agents SDK — the framework for building AI agents — now includes a LiteLLM integration supporting 100+ rival models. OpenAI built official plumbing to replace OpenAI.

Who Made Your Model?

The pattern is clear and weird. Cursor's crown jewel runs on a Chinese base model. Microsoft distributes Anthropic through the product it funded with OpenAI money. OpenAI supplies the escape hatch from itself.

This isn't multi-model freedom. This is supply chain fragmentation.

When Composer 2 generates code from Kimi K2.5, who's accountable if the output contains patterns from the base model's training data? Cursor, who fine-tuned it? Moonshot AI, who built it? Nobody has decided. The dropdown gives you a choice of engines without a map of where those engines came from or what they were trained on.

Six months ago, your vendor chose the model, which meant your vendor owned the liability. Now you pick from a menu, and the accountability chain splits across companies, countries, and training datasets you'll never see.

The Brand Collapse

The vendor logo on your code editor used to be an engineering claim. "We built the AI" meant something about quality control, training data curation, and safety testing. That claim evaporated this month.

What replaced it is retail. Cursor is a store stocking models from San Francisco and Beijing. Copilot is a storefront where Microsoft shelves its competitor's product next to its partner's. The brand now means UX and integration depth — not model origin.

What This Actually Means

The model dropdown didn't give you freedom. It handed you a supply chain you're now responsible for understanding. Which model handles your proprietary code? Where was it trained? Who fine-tuned it, and on what data?

Those used to be the vendor's questions. In April 2026, they became yours.

Open your code editor. Look at that dropdown. You're no longer picking a model — you're picking a supply chain. And nobody shipped a label for what's inside.