Meta AI Updates: May 17, 2026

1. Threads users revolt after discovering the @meta.ai account cannot be blocked

Meta. The @meta.ai Threads beta that landed in five pilot markets has escalated into a platform-wide backlash, with the hashtag tracking “Users cannot block Meta AI” surging past a million posts and users reporting the account for spam only to find no block ever takes effect. Meta’s spokesperson told Engadget that users can mute the account or hit “not interested” to see fewer of its replies, but the company is keeping the bot unblockable across the rollout. The episode is a useful case study for any team shipping an always-on AI surface inside a social product without a hard opt-out. Source

2. Threads AI backlash hardens around the missing opt-out

Meta. Dataconomy’s deeper write-up details how the Threads AI account, currently limited to Malaysia, Saudi Arabia, Mexico, Argentina, and Singapore, surfaces public replies tagged with @meta.ai but bypasses the standard Threads block stack entirely. The piece breaks down why this matters for the underlying moderation model: Meta is treating @meta.ai more like a system account than a user account, which collapses the usual user-versus-user trust controls. The framing lines up with concerns also raised in Digital Trends and Gadget Review coverage from the same window. Source

3. Meta opens Ray-Ban Display smart glasses to third-party developers

Meta. Technobezz reports that Meta has officially opened the Ray-Ban Display smart glasses to third-party developers and apps, expanding the on-device app surface beyond Meta’s first-party Llama-powered assistant. The move pushes the device toward becoming a Llama-backed wearable platform rather than a single-app accessory, with hands-free voice and on-lens display surfaces now exposed via developer APIs. For builders, it is the first concrete signal that the Ray-Ban Display SDK is shipping outside Meta’s labs. Source

4. Meta PyTorch team publishes H1 2026 roadmap on the dev mailing list

Meta. The Meta PyTorch team posted its H1 2026 roadmap on the official developer mailing list, laying out the half-year priorities the group plans to globally optimize across Meta’s internal users and the wider PyTorch community. The thread is the canonical place for downstream framework users to see what Meta is committing engineering capacity to in the first half of 2026, including ongoing compiler, export, and on-device inference work that has been visible across the 2.12 release cycle. Useful primary-source reading for anyone tracking the framework’s direction. Source