Threads adds beta of public Meta AI assistant that replies to mentions: what changed

Meta has begun a Threads beta that lets users mention Meta AI to get public replies from the @meta.

Meta has started a beta on Threads that lets users summon an on-platform AI assistant by mentioning Meta AI in a post or reply, and the assistant responds publicly as the @meta.ai account. The move aims to provide real-time context about trending topics and breaking stories inside the app rather than directing users to external searches, a shift that could make factual and contextual answers more visible in public conversations.

The capability is currently limited to public accounts and is available in Malaysia, Saudi Arabia, Mexico, Argentina and Singapore. When invoked, the system processes the mention and posts a public reply authored by @meta.ai rather than producing an inline private response; it answers in the language used in the original post.

Meta described the integration as intended to surface recommendations and immediate context directly within conversations. Example prompts the company highlighted include questions such as “Why are people talking about the World Cup this month?,” “Whose Met Gala looks are trending right now?,” and “How are the Knicks doing in the playoffs?”. These examples illustrate the assistant’s focus on trends, cultural moments and sports updates.

By replying publicly as @meta.ai, Threads positions the feature as a hybrid between a social feed and a conversational knowledge surface where AI-generated replies appear alongside ordinary posts and replies in the public timeline. That design mirrors the role Grok plays on X, where users often tag the chatbot to verify claims or request explanations, and reflects a broader industry trend of embedding visible AI assistants to increase engagement.

Visible AI replies carry safety and reputational trade — offs. Reporting cites prior incidents with Grok on X-including an instance in which the chatbot produced praise for Hitler — as an example of how automated outputs can cause harm when surfaced broadly. Meta says its safeguards are stronger than Grok’s but acknowledges that real-world moderation and failure modes will be tested during this beta.

To reduce exposure, Meta has built user controls that let people mute the @meta.ai account, mark a Meta AI post as “Not interested,” or hide a Meta AI reply that appears on their post. The company told reporters it will collect early feedback and iterate on the experience before expanding access, leaving specific timing and broader rollout plans unspecified.