Meta Is Testing Community Notes Across 16 Latin American Countries Before the U.S. Version Has Been Audited

On January 7, 2025, Meta said it would wind down its third-party fact-checking program in the United States and move Facebook, Instagram, and Threads toward a Community Notes model popularized on X. Mark Zuckerberg framed the old system as a tool that too often became censorship. Newsrooms that had staffed the contracts framed the announcement as an abdication timed to a new administration. Both descriptions can be true. The product question was always narrower: who writes the note, who rates it, and how far a false claim travels before anything attaches. In September 2026 Meta answered a different question, which is where it is willing to try the experiment next, and it answered before anyone outside the company can audit whether the first market works.

Meta’s own transparency page, updated September 9, 2026, says Community Notes is fully launched only in the United States. The company is beginning to test the feature in sixteen other countries: Argentina, Bolivia, Chile, Colombia, Costa Rica, Ecuador, El Salvador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay, and Venezuela. People there can sign up as contributors. Notes will not be published, and the third-party fact-checking program will not be switched off, until Meta says it is confident the product can succeed. That is a narrower step than a global replacement. It is still a step the company’s own Oversight Board asked it not to take on a timetable this thin.

The Board Asked for a Report It Says It Has Not Received

On March 26, 2026, the Oversight Board issued a policy advisory opinion on expansion beyond the United States. The Board said notes can improve discourse if they arrive with speed, scale, and protection against manipulation. It also said a one-size rollout can cause real-world harm in repressive states, in places with complex political violence, in countries with a history of large coordinated disinformation networks, and in election windows and active crises. It recommended that Meta not treat notes as a primary safeguard in countries in crisis, that it delay launches where a language community is too thin to rate notes, and that it build a country-level risk matrix and show the Board how that matrix is applied.

On September 10, 2026, the Board’s co-chairs answered the Latin America test directly. They said they were watching the sixteen-country trial, and they listed three things they still wanted. The decision to move from testing to implementation should be driven by country-specific risk, and the Board said it had not yet received the first report on the criteria Meta intends to use. Any expansion should wait on a judicious review of how the program is performing in the United States, with researcher access to the data. And notes should not be treated as a simple replacement for professional fact-checking. Meta is not required to obey the Board. Historically it has accepted a majority of recommendations and ignored the rest. Readers should treat the September statement as a record of a request that was still outstanding on the day the test began, not as a veto.

What the United States Version Has Actually Published

The American record, which is the only full launch Meta has, does not look like a system that has earned a victory lap. By September 2025, Meta executives were telling audiences that only about six percent of proposed notes ever reached publication. In the first six months of the U.S. rollout, the company’s chief information security officer said roughly nine hundred notes had been published. European fact-checking partners, still working under a different set of rules, pointed to tens of millions of labels on Facebook in the European Union over a comparable window. Those figures are not a perfect comparison. They are close enough to show a scale problem. A system that publishes hundreds of notes cannot sit in front of a platform that hosts billions of items unless the notes attach to the posts that actually move.

Academic work on X’s older version of the same idea has not been flattering on speed. Proposed notes often sit in a “needs more ratings” pile for many hours, sometimes days, and a large share never publish. The Los Angeles fire season that followed Zuckerberg’s announcement was an early, ugly test on Meta’s own apps. Conspiracy posts moved at the speed of smoke. Notes arrived later, if they arrived. The BBC asked the obvious question weeks afterward: can a crowdsourced label scale without becoming another faction? The European Federation of Fact-Checking Organizations called Community Notes inadequate as a standalone solution and urged a hybrid. That is a trade association talking its book. It is also describing a gap Meta’s own six-percent figure already admits. A hybrid would keep professional checkers for crisis windows and let notes handle the long tail of checkable junk. The January 2025 announcement did not offer a hybrid in the United States. It offered a replacement. The September 2026 test offers a waiting room, not a hybrid.

Brazil Is the Name Missing From the List

The sixteen-country list is almost entirely Spanish-speaking Latin America. Brazil, the region’s largest market for these apps, is not on it. Coverage of the announcement, including a report from El Economista, noted that Brazil was left off while a presidential election sat on the October calendar. Meta’s public explainer does not give a written reason. The omission can be read as caution, which would match the Board’s warning about election windows. It can also be read as a decision to test in markets where a thin contributor pool is easier to stand up than in a Portuguese-language electorate of that size. The AEGIS Alliance will not pretend to know which reading Meta prefers until the company publishes the risk criteria the Board says it asked for. What can be said now is that a test which includes Venezuela, Nicaragua, and El Salvador, and which skips Brazil, is not a neutral sample of “Latin America.” It is a list. Lists have politics even when the product team calls them a beta.

The honest arguments from 2025 still sit on the same page. One is that Meta waited until the political cost of fact-checking exceeded the reputational cost of dropping it, and that Donald Trump’s praise of the move was not a coincidence. The other is that a small set of contracted newsrooms had been given a throttle on distribution and used it in ways that looked like faction work. Neither argument answers the product question. A false medical claim, a forged campaign video, or a rumor that sends people toward a disaster site does not become less dangerous because the old checkers were unpopular. It also does not become more true because a crowd of raters failed to reach consensus before the clip peaked. Handing part of that job to users in San Salvador or Caracas, while fact-checkers are still on contract there, is an experiment. Calling the experiment a speech renaissance is advertising.

How to Read a Note, and How to Read a Test

A Community Note is a claim about a claim. It is only as good as the people who can write it and the people who can rate it before the original post is done traveling. If the note is late, it is a correction for the archive and a failure for the people who already acted. If the note is captured by a faction, it is a laundered pile-on. If no note appears, the platform has chosen silence and called it humility. In the sixteen test countries, the additional risk is that a signup drive creates the appearance of a safeguard before a single note is allowed to publish. Contributors can be recruited. The old fact-checking contracts can keep running. The public can be told that a transition is underway. That is a communications product as much as a moderation product.

Adjacent files on this desk belong next to this one for a reason. TikTok’s forced ownership fight is about who sits between a speaker and an audience when a statute does the sitting. The verdict that held Meta and Google liable in a social-media addiction case is about what a courtroom does when a product argument becomes a harm argument. An older piece on this site, Facebook’s expectation that pages would fact-check themselves, shows how long this company has been trying to push the work outward. Readers who want the technology file can start in the technology section.

The AEGIS Alliance will keep watching three numbers: the U.S. publication rate, the time it takes a note to reach consensus, and whether the missing risk report ever arrives before any of the sixteen countries is allowed to publish. Zuckerberg got the political weather he wanted in January 2025. In September 2026 the company started recruiting contributors in places the Board had flagged, while telling the public the old checkers are still on duty there. That is not the death of truth, and it is not a renaissance of speech. It is a rollout that got ahead of its own audit. The notes, when they finally appear, should be read as arguments. The test should be read as a choice about whose argument is allowed to arrive on time.