Apple engineers have quietly built an internal, ChatGPT-like prototype to probe how far Siri can be pushed, according to a Bloomberg report that offers a rare glimpse into the company’s behind-the-scenes artificial intelligence work. The prototype, described by Bloomberg as a testing platform rather than a customer-facing product, was one step in a broader strategy that Apple now plans to realize in 2026 by combining its own models with at least one third-party offering.

Bloomberg says the company looked earlier this year at models from OpenAI and Anthropic, and by August was “circling” a partnership with Google — a move that would underscore how even dominant hardware makers are weighing reliance on competitor technology to close capability gaps. Apple has not publicly disclosed the details; the company typically declines to comment on rumors, and a spokesperson did not respond to a request for confirmation.

The effort marks a significant pivot for Apple, which has historically emphasized owning the full stack — silicon, software and services — as a competitive advantage. Siri, introduced in 2011, has lagged in generative conversational abilities compared with chatbots such as OpenAI’s ChatGPT and Google’s Bard. Combining on-device models tuned for privacy with cloud-based models from others could accelerate Siri’s conversational fluency, expand its knowledge and allow richer, context-aware interactions across iPhones, iPads and Macs.

But the approach raises thorny trade-offs. Privacy advocates and regulators have long scrutinized how companies route personal data to third-party services; handing user queries to external large language models would require careful architectural and contractual safeguards if Apple seeks to preserve its privacy-first branding. The decision to mix proprietary and partner models also touches antitrust sensitivities: a partnership with Google, the dominant search and cloud provider, would invite renewed questions about competitive influence inside Apple’s tightly controlled ecosystem.

AI researchers say the hybrid model strategy is pragmatic. “Teams often build simplified internal versions of complex systems to stress-test integration and edge cases,” said an AI developer briefed on industry practices. Such prototypes allow engineers to measure latency, hallucination rates and the practical limits of on-device inference before committing to a particular external supplier.

For consumers, the practical outcome could be a more useful Siri that handles complicated follow-ups, composes longer responses and better integrates across apps. That promise will likely be paired with monetization choices, such as premium subscriptions for advanced conversational features or tighter hardware-device integration that leverages Apple’s neural engines.

Apple’s timeline — public reports point to a 2026 rollout — gives regulators and competitors time to adapt. It also signals how generative AI has become not just a software problem, but a strategic business decision involving partnerships, privacy guarantees and hardware investments. As the industry braces for the next wave of consumer-facing AI, Apple’s move to prototype an in-house ChatGPT-style model highlights both the urgency and complexity of catching up in an era defined by language models.