Apple tests internal "Apple GPT" chatbot

While OpenAI, Microsoft, and Google have spent the past several months dominating the headlines, Apple has been remarkably quiet. As it turns out, the company hasn't been ignoring the generative AI craze—it has just been testing it behind tightly closed doors.

According to a report from Bloomberg's Mark Gurman, Apple is building its own artificial intelligence tools, including a chatbot that engineers internally refer to as "Apple GPT."

Under the hood, Apple built a unified foundation codenamed "Ajax" to create large language models. Interestingly, Ajax runs on Google Cloud and was constructed using Google's machine learning framework, JAX. Apple is using this framework to train models and power its internal conversational tool.

Strictly locked inside Cupertino

The internal rollout hasn't been entirely smooth sailing.

Apple initially halted deployment inside the company due to security concerns around generative AI. While access has since expanded to more staff, it still requires special executive sign-off.

There is also a strict guardrail in place: Bloomberg reports that any output generated by the chatbot cannot be used to build customer-facing features. For now, staff are using it primarily for product prototyping, summarizing text, and answering questions based on training data.

Functionally, it doesn't pack any secret sauce just yet. Those with access say it is essentially a replica of ChatGPT, Bard, and Bing AI, lacking distinct consumer-ready features.

Behind the scenes, Apple is on a hiring spree for engineers with deep expertise in large language models. Apple CEO Tim Cook has stated the company plans to integrate AI on a "thoughtful basis," and sources suggest a major AI announcement could land next year.

Why it matters

For everyday Apple users, don't expect a supercharged Siri on your iPhone tomorrow morning. This tool is locked in Cupertino, and consumers cannot test it yet.

In our view, Apple's slow-and-steady approach is very deliberate. While Microsoft and Google race to push experimental models into production, Apple has brand reputation and user privacy to protect. Hallucinations and data leakage are risks Apple simply cannot afford on over a billion active devices.

Leaning on Google Cloud and Google JAX to spin up Ajax is a practical shortcut. Instead of reinventing machine learning infrastructure from scratch, Apple is catching up on research speed.

The real challenge will be execution. If the underlying chatbot is currently no different from standard commercial alternatives, Apple must figure out how to bring these capabilities to consumer hardware without violating its strict privacy stance. We expect next year will reveal whether Apple can truly leapfrog the field or if it is merely playing defense.