Vercel AI SDK patch changes validation and batch-output handling
Version 7.0.174 includes a documented change intended to reduce resource use when validating deeply nested metadata.

The Vercel AI SDK's ai package release 7.0.174, published on GitHub on 17 November, includes changes to batch outputs, streaming behavior and message validation. The maintainers say deeply nested provider and tool metadata could previously require excessive memory and CPU during validation.
According to the release notes, the affected validation now uses a single pass and avoids copying values except when removing __proto__ keys. Other listed changes include support for output in startBatch and parsing it in getBatchResults, plus a fix for transforms receiving an undefined tools value when no tools were supplied. These details matter to developers integrating the package rather than to ordinary chatbot users.
The release note does not establish that a particular deployed app was attacked or that every SDK user experienced the issue. Teams should check the version they actually ship, review the changelog and run their own streaming, batch and validation tests before updating production.
Sources
Source: Vercel AI SDK GitHub release notes ↗
Updated: 20/11/2026, 10:39:08 am IST



