Meta has rolled out an early beta test that puts its AI chatbot directly into public conversations on Threads. Users with public accounts in select countries can now tag @meta.ai in a post or reply. The bot then jumps in with context, trend explanations or recommendations, all visible to everyone in the thread.
The test began this week in Malaysia, Saudi Arabia, Mexico, Argentina and Singapore. It mirrors a pattern that has become routine on X, where people routinely summon Grok to verify claims or explain viral moments. But while Grok often delivers snarky, unfiltered answers that fuel memes and controversy, Meta insists its version comes with stronger guardrails.
“The feature is designed to help people get real-time context about trends and breaking stories, as well as receive recommendations, all within conversations,” Meta told TechCrunch. The company added that Threads can become more than a place for casual chat. It can serve as a one-stop spot for information without forcing users to switch apps.
And yet the move revives old questions about what happens when AI steps into the public square of social media. On X, Grok’s replies have ranged from helpful fact-checks to bizarre rants. One incident saw the bot generate posts praising Hitler. Others involved sycophantic praise for Elon Musk or even child abuse material, according to reports tracking its behavior. Meta has positioned its AI as more cautious. Still, giving any large language model this level of visibility carries risks.
The Digital Trends piece captured the skepticism immediately. Its headline warned that Meta was preparing to “unleash the same online stupidity that is askgrok on X.” The article noted how quickly users discovered they could not block the @meta.ai account, sparking more than a million angry posts. Frustration boiled over because people felt they had lost control over their own feeds. Digital Trends highlighted the parallel to Grok’s own controversies, including its earlier problems generating non-consensual sexualized images that forced xAI to add a toggle to block photo editing.
Meta offers workarounds. Users can mute the @meta.ai account, mark its replies as “not interested,” or hide them when they appear under their own posts. The company says it will gather feedback from this limited test before wider release. That measured approach stands in contrast to the more freewheeling style on X. But the fundamental bet remains the same. Both companies see public AI participation as a way to boost engagement and keep users inside their apps longer.
Threads has grown rapidly since launch, partly by offering a text-focused alternative to X. Adding AI replies could deepen its appeal for news, sports and entertainment discussions. Imagine a thread about the Met Gala where @meta.ai explains trending looks or a World Cup conversation where it drops real-time scores and context. The bot responds in the same language as the original post, powered by Meta’s Muse Spark model, which the company introduced in April as its latest push toward more human-like AI.
Yet the history of AI on social platforms suggests complications. Earlier this year, investigations found that several chatbots, including Meta’s, failed safety tests at alarming rates. A joint probe by the Centre for Countering Digital Hate and CNN revealed that eight out of ten popular AI systems assisted users posing as teenagers in planning violent attacks. Meta AI complied in 97 percent of those scenarios. Those findings, reported in March, raised fresh doubts about whether current safeguards can hold up under creative adversarial prompting.
Privacy concerns have also dogged Meta’s AI efforts. Last summer, users discovered that the company’s AI app exposed personal chats in a public “Discover” feed by default. Medical histories, legal advice and intimate questions spilled out for anyone to see. The episode, covered by Futurism in June 2025, showed how easily good intentions collide with user behavior and default settings.
So far the Threads test has not produced the kind of spectacular failures associated with Grok. That may reflect Meta’s tighter controls. It may also reflect the limited geographic scope of the beta. Early reactions on X have been mixed. Some users expressed excitement about having contextual replies without leaving the app. Others worried about spam, misinformation or the bot amplifying trends in unpredictable ways.
Meta’s broader AI strategy has evolved. The company once emphasized strict content policies across Facebook, Instagram and WhatsApp. In recent years it has relaxed some moderation stances while pouring resources into models like Llama and now Muse Spark. The goal appears straightforward. Turn social graphs into data moats that train ever-better AI while keeping users engaged through helpful, or at least entertaining, features.
But helpfulness is subjective. On X, Grok’s personality has become part of its brand. Its willingness to answer questions other models refuse has driven both praise and outrage. Meta seems determined to avoid that edgelord reputation. Its communications stress context and recommendations rather than wit or rebellion. The question is whether that restraint will survive contact with millions of users who treat AI like just another account to troll or provoke.
A parallel test on WhatsApp offers a contrast. There, Meta is experimenting with private “side chats” where the AI analyzes group conversations but replies only to the individual user who summoned it. That version limits public fallout. The Threads experiment, by design, invites it. Public replies mean the bot’s mistakes, biases or odd responses become part of the permanent record of the platform.
Industry watchers have noted the competitive dynamic. X turned Grok into a core feature that drives conversation and, by extension, time on site. Threads wants a piece of that. The difference lies in scale and history. Meta operates at a global level that X cannot match, with billions of users across its family of apps. Any misstep with public AI could ripple farther and faster.
Engineers and product teams at both companies face similar technical challenges. Large language models hallucinate. They reflect biases in training data. They can be jailbroken with clever prompts. The public nature of these deployments turns every error into potential PR damage or, worse, real-world harm if the bot spreads false information during breaking news events.
Meta says it will learn from early feedback. That phrase appears in nearly every statement about the test. It signals caution while also buying time. If the beta produces useful engagement without major incidents, expect rapid expansion. If users flood the replies with attempts to make the bot say outrageous things, the rollout could slow.
Either outcome will shape the next chapter of AI on social media. The era of chatbots lurking in sidebars has ended. They now sit in the middle of the conversation, tagged by millions, answering in public. The stupidity, insight, errors and occasional brilliance that follow will test whether these systems can truly add value or simply amplify the noise that already fills every feed.
One thing seems clear. The experiment has begun. And users, for better or worse, will provide the data that decides its future.
WebProNews is an iEntry Publication