Accessory Access: IOUSBHostInterface can’t be opened after AAUSBAccessory.open() (QEMU-based VM app)

Hi,

I’m building a macOS hypervisor app based on QEMU + Hypervisor.framework (a fork of the open-source UTM project). Following Quinn’s advice in DTS case 23599042, I’m trying to implement USB passthrough on macOS 27 with Accessory Access instead of com.apple.vm.device-access. Since the app doesn’t use the Virtualization framework, I take the IOUSBHostDevice from AAUSBAccessory.open(serviceQueue:) and forward the transfers to QEMU through usbredir.

What works

“Claim USB Accessory” capability (com.apple.developer.accessory-access.usb), plus com.apple.security.app-sandbox and com.apple.security.device.usb
registerListener(:matchingCriteria:); the user assigns the device from the “Virtual Machine Accessories” menu bar item; usbAccessoryDidConnect fires
open(serviceQueue:) returns an IOUSBHostDevice; control transfers with enqueueDeviceRequest work; configureWithValue(
:matchInterfaces:) succeeds

What fails
Opening any interface. IOUSBHostInterface(__ioService:options:queue:interestHandler:) on the device’s child IOUSBHostInterface services always fails:

Error: Unable to open io_service_t object and create user client. with reason: IOServiceOpen failed.
Error: Failed to create IOUSBHostInterface. with reason: Failed [super init]
IOUSBHostErrorDomain 0xe00002c9

I tried with the interfaces as delivered, after configureWithValue(1, matchInterfaces: false) and after configureWithValue(1, matchInterfaces: true). No sandbox violations are logged. Device: USB 3 mass storage stick. macOS 27.0.1, Xcode 27.2.

Questions

Is interface and pipe access (IOUSBHostInterface / IOUSBHostPipe) supposed to work for an AAUSBAccessory opened by an app that doesn’t use the Virtualization framework?
If so, what’s the right way to get those objects?
If not, is the default control endpoint the intended scope? In that case I’ll file a feedback report, since bulk and interrupt endpoints are needed for USB passthrough in non-Virtualization hypervisors.

Thanks!