Meta AI is now summarizing wild Facebook comments

Meta is now summarizing Facebook comment sections with AI, and the results are just as unhinged as you'd expect.

Facebook comments have never been known for nuanced discourse. Naturally, Meta thought it was a great idea to park its AI at the top of the thread to distill the collective chaos into neat little bullet points.

AI tries to make sense of the internet

Screenshots of the feature started surfacing on Threads and Reddit before reporters at The Verge spotted them live in the wild.

On a post about a store closing down, Meta AI dutifully explained that some blamed the closure on the store "going woke" while others cited the rise of e-commerce. On a Vice post covering Mexican street wrestlers, the bot noted that some viewers were less than thrilled, branding the act a "moronic way of panhandling." And on a clip of a bobcat in Florida, the AI earnestly highlighted a comment hoping the animal remembered to wear sunscreen.

It is still unclear what triggers the AI summaries to appear on specific posts. Meta did not respond to requests for comment.

This is not the first time Meta AI has popped up uninvited in conversations. 404 Media previously caught the bot inside a Facebook group, bizarrely pretending to have a child of its own.

The data harvesting problem

Beyond being unintentionally hilarious, the feature is driving renewed scrutiny around how Meta feeds its models.

Over the past week, users across the European Union and the UK received notices informing them that Meta will use their public posts and comments to train its AI. While EU privacy laws require Meta to provide an opt-out form, users report the process is surprisingly difficult, and Meta has already denied several requests.

In the U.S., the rules are far looser. Meta's privacy policy makes it clear that virtually everything posted on its services, including captions, photos, and comment rants, is fair game for model training unless it originates from a third-party broker.

Why it matters

We think this feature is a prime example of shipping AI for the sake of shipping AI.

Nobody goes to a Facebook comment section looking for an executive briefing. People go for raw reactions, jokes, or drama. Having a bot sanitize and regurgitate random hot takes adds virtually zero value to the user experience, except maybe serving as a quick warning label to avoid toxic threads.

More importantly, it puts Meta's data appetite on full display. Every stray remark and internet squabble is being ingested into Meta's massive LLM pipeline. While Europe is putting up modest legal resistance, U.S. users are essentially powerless against their digital footprints being absorbed into corporate model weights.