Apple built an internal ChatGPT app to overhaul Siri

Apple just built an internal ChatGPT clone to fix Siri.

According to a Bloomberg report, Apple quietly developed a chatbot app called Veritas. It is not for you or me, but for Apple engineers trying to salvage the delayed Siri overhaul.

A secret sandbox named Veritas

Veritas looks and acts like any popular AI chatbot. Testers can manage multiple threads, save chat histories, and go back and forth across long prompts.

Engineers are using it to put the overhauled Siri through its paces. They test personal data lookups, web summarization, and in-app actions like photo editing.

Under the hood sits Linwood, Apple's new backend system. Linwood combines Apple's in-house Foundation Models with third-party LLMs. After talks with OpenAI and Anthropic, Apple recently ramped up discussions with Google to tap into Gemini.

One-third failure rates and executive shakeups

Apple originally intended to roll out this revamped Siri earlier this year. But the code kept breaking.

In testing, new features broke in up to one-third of trials. That forced Apple to delay the target launch to March 2026 at the earliest.

The fallout has already hit leadership. Robby Walker, who previously oversaw Siri, is leaving Apple in October. Meanwhile, AI chief John Giannandrea and his lieutenants have been sidelined, with software boss Craig Federighi taking the wheel.

Tim Cook reportedly told employees that AI is the biggest transformation in decades and insisted Apple must win. Behind the scenes, the company is also working on a redesigned assistant interface for late 2026 and bringing AI to HomePods and Apple TVs.

Why it matters

Apple admitting it needs a chatbot sandbox to test Siri shows just how messy its AI pivot has been.

For users, the takeaway is clear: don't hold your breath for a genius Siri on your next phone. The real overhaul is pushed to at least March 2026, meaning everyday Apple fans are stuck with a clumsy assistant while Google and Samsung pull ahead.

Industry-wise, leaning into Google Gemini proves Apple cannot pull this off alone. A 33% failure rate in testing is brutal for a company built on polished hardware-software integration.

Putting Craig Federighi in charge makes sense. The academic approach under Giannandrea fell behind, and Apple now needs a pragmatic software chief who can ship stable features, even if it means relying on competitors' models.