51吃瓜

On March 26, Meta鈥檚 Oversight Board a warning that should end the idea that crowdsourced correction programs, as currently designed, can keep pace with viral falsehoods. In a policy advisory opinion requested by Meta, the Board said Community Notes can help only if they have enough scale, speed and safeguards against manipulation. It added that publication delays, the small share of notes that ever appear and dependence on the surrounding information environment raise serious doubts about whether the system can meaningfully curb harmful misinformation, especially in 鈥渞epressive human rights regimes, in particular electoral contexts, and in ongoing crisis and conflict situations.鈥�

That policy debate can sound abstract. It is not. In Ethiopia, Professor Meareg Amare was in social media posts that identified him by name, photograph, workplace and home address, and falsely accused him of backing the Tigrayan People鈥檚 Liberation Front. His son, Abrham, later said he understood immediately that the posts were a death sentence. Amnesty International that platform failures to adequately moderate content contributed to serious abuses against Tigrayans during the war in northern Ethiopia.

That warning matters because the medium is increasingly video. By the end of 2025, video accounted for 76% of all mobile data traffic, with social media video comprising % of smartphone video traffic in sampled European networks. Video is potent because it stacks image, voice, music, captions, pace and the seeming intimacy of a familiar face into one persuasive package. Research on parasocial relationships suggests that one-sided familiarity with a creator can heighten trust and perceived credibility, making audiences more receptive to that person鈥檚 claims. That is excellent for education, journalism and public communication. It is equally useful for manipulation.

Video has become one of the world鈥檚 dominant forms of public persuasion, yet verification still lags behind. With limited exceptions, fact-checking has been too slow, too patchy and too far removed from the viewing experience.

The human cost of unchecked misinformation

Meanwhile, the consequences keep recurring. In April 2024, after the stabbing of Bishop Mar Mari Emmanuel during a livestreamed church service in Wakeley, Australia, rumors raced through WhatsApp while the video bounced across phones; within hours, police officers were injured in the riot that followed. In Britain that summer, false claims about the Southport attacker 鈥� especially the lie that Axel Rudakubana, the teen convicted of killing three girls at a Taylor Swift-themed dance class, was a Muslim asylum seeker 鈥� helped fuel riots across England and Northern Ireland, targeting communities with no connection to the crime. A society need not be poor, fragile or far away for this pattern to take hold. It needs grievance, speed and a public primed to experience video as proof.

These harms also do not fall evenly. Especially in countries already in conflict or crisis, the spread of misinformation and disinformation can intensify violence, acts of discrimination, and abuse against human rights defenders, racial and religious minorities, women politicians, humanitarian workers and others. But governments too often justify the of free expression 鈥� and, increasingly, the shutting down of the internet 鈥� by citing misinformation and 鈥渇ake news鈥� when they wish to silence criticism or other opinions they find politically problematic. It鈥檚 vital that standards and systems established to address harmful misinformation be rooted in respect for freedom of opinion and expression as defined in international human rights law.

Platforms鈥� failures and limitations

That same human rights lens should also be applied to the platform systems designed to respond to misinformation. Meta鈥檚 Oversight Board warned that coordinated networks can game Community Notes, that the system can privilege dominant groups over minorities, and that it should not be introduced in crisis or protracted conflict conditions.

And the empirical record for Community Notes on X is sobering as well. A 2025 Digital Democracy Institute of the Americas (DDIA) of X鈥檚 full public dataset found that more than 90% of submitted notes never reached the public. In English, only 7.1% of submissions from January 2021 through March 2025 were published, falling to 4.9% in early 2025. Average publication time in 2025 was 14 days. A Washington Post found that among election-related false or misleading posts where users had already written accurate, relevant notes, 91% never became visible.

Recent evidence further sharpens the timing problem. A 2025 Proceedings of the National Academy of Sciences found that Community Notes attached within one to 12 hours reduced repost growth by 49.6%. But once notes slipped into the roughly two-day range, the effect on repost growth fell to 6.2%, and the total repost reduction across a post鈥檚 life was effectively zero.

A path forward: real-time verification

The big question, then, is whether context can appear while the video is still shaping perception. It can now. Technology such as offers one example. This browser extension overlays real-time verifications like subtitles, checks claims against depersonalized web searches and integrates factual reliability scores from independent media-rating agencies. The broader point is not about any one application; it鈥檚 that the technology now exists to meet viral claims with contextual verification at the speed and scale the medium demands.

This matters for human rights, public safety and countries that still imagine themselves insulated from these risks. False accusations carried by video can inflame a neighborhood in Sydney, a city in England, a borderland in East Africa and spread through a diaspora community worldwide. They can trigger a mob against immigrants, a retaliatory attack on a religious minority or a crackdown later justified as restoring order. The pattern endures because the architecture of outrage does; velocity first, questions later, if at all.

Our societies should treat safer information infrastructure as we鈥檝e learned to treat smoke detection, emergency alerts or fraud monitoring 鈥� as essential. An information safety layer must work in real time, across languages and before the damage compounds. People must still be free to speak, document, argue and dissent. But they should also be free to choose third-party software that can add context immediately, when a falsehood is still taking hold.

That is the standard by which this debate should now be judged: whether verification can appear before a viral falsehood hardens into a weapon. We now have a choice. Technology exists to address social media misinformation for people and institutions who want to do so today. We can keep accepting an information order in which viral falsehoods outrun verification, or we can adopt one in which facts can outrun misinformation, while still protecting lives and public order.

[Iain Levine is a consultant on human rights and was previously a director on Meta鈥檚 human rights team. Avi Tuschman is a Stanford StartX entrepreneur and the founder of .]

[ edited this piece.]

The views expressed in this article are the author鈥檚 own and do not necessarily reflect 51吃瓜鈥檚 editorial policy.