Apple AI Updates: May 20, 2026

1. Apple Intelligence powers a wave of new accessibility features ahead of WWDC

Apple. In a Global Accessibility Awareness Day push on May 19, Apple announced Apple Intelligence-backed upgrades across VoiceOver, Magnifier, Voice Control, and Accessibility Reader, with VoiceOver image recognition now reading detailed structured fields out of photos (an Apple example: pulling the amount and due date directly off a bill) and a new natural-language layer letting users say things like “Tap the guide about best restaurants” in Apple Maps to take action on what they see. Apple is also rolling out AI-generated subtitles for videos without pre-baked captions, including clips recorded on iPhone or received in Messages, plus a power-wheelchair control mode for Apple Vision Pro. The features ship as free updates across iOS 26, iPadOS 26, macOS Tahoe 26, watchOS 26, and visionOS 26 later this year, and double as a preview of how Apple’s on-device AI stack is being positioned in the weeks before WWDC on June 8. Source

2. Apple details how MLX uses the M5 Neural Accelerators for on-device LLMs

Apple. A new Apple Machine Learning Research post walks through running large language models with MLX on the M5 GPU’s Neural Accelerators, with benchmarks for time-to-first-token and decode throughput on the new 14-inch MacBook Pro and a requirement of macOS 26.2 or later to unlock the accelerated path. The piece is Apple’s first concrete public guidance on the M5 generation’s matmul units for transformer inference and reinforces MLX as the recommended framework for local LLM work on Apple silicon, including for partners like Ollama that have already wired MLX into their Mac builds. For developers, the practical takeaway is that the M5 path is finally documented enough to evaluate against CUDA-class boxes for small-batch local inference. Source