केस विवरण
The Oversight Board has accepted Meta’s request for a policy advisory opinion on its approach to expanding its community notes program outside of the United States.
Meta has requested the Board’s guidance on the factors it should consider when deciding whether any country should be omitted from its community notes expansion, as contextual elements may impact the program’s operations. Additionally, Meta has asked the Board how to weigh those factors in relation to one another, in a way that can be applied on a large scale.
In its request, Meta said that the community notes program is in an “early stage of product development” and it possesses “limited data from the US beta rollout.” Because of these considerations, the company’s “primary interest lies in establishing fundamental guiding principles” for its rollout worldwide.
On January 7, 2025, Meta announced that it was ending its third-party fact-checking program in the United States and transitioning to community notes. At the time, Meta indicated that it would refine community notes before making it available to users outside the United States. Community notes allows users to add labels with additional context to potentially misleading content (in contrast to the third-party fact-checking program, which relies on partner organizations to label misleading information).
Meta’s request to the Board is here.
In its request, Meta describes how community notes work. Meta users apply to contribute to the program. Should they meet Meta’s eligibility criteria, they are “gradually and randomly” admitted from the waitlist and may then write and rate notes. At present, contributors can compose and submit notes to “add more context” to public, organic content on Facebook, Instagram, and Threads originating in the United States. They have access to a dedicated feed of posts that users have flagged as having the potential to benefit from a note. Contributors must include a link supporting the context shared in the note. They also have the option to rate notes written by other contributors as “helpful” or “not helpful” and explain their response by selecting a reason from a list of options.
Meta disclosed that it built its community notes system using the open-source algorithm from the community notes program of social media platform X. Meta described the algorithm as a “consensus algorithm that uses separate measures of ‘helpfulness’ and ‘consensus’ to calculate an overall ‘helpful consensus’ score.”
In its request, Meta states that the algorithm calculates this score by identifying agreement that a note is helpful among a sufficient number of contributors who usually disagree with each other based on past ratings. According to Meta, if the combined “helpful consensus” score on a note exceeds a “certain threshold” and the note does not violate Meta’s Community Standards, the note will be published. The note appears as a banner on the bottom of the underlying post, which users can click to read the full note and supporting link. Meta has said this approach “helps ensure that notes reflect a range of perspectives and reduces the risk of bias.”
The request describes steps Meta is currently taking to retain and support volunteer community notes contributors, as well as to prevent coordinated manipulation of submission and rating of notes.
Meta states that its approach to enforcing its Misinformation and Harm Community Standard remains unchanged. Under this policy, the company removes misinformation where it is likely to directly contribute to the “risk of imminent physical harm” and “interference with the functioning of political processes.” Meta continues to use trusted third parties to help identify content violating that community standard.
Meta’s questions to the Board:
Meta presented the following list of factors it might consider when deciding which countries to omit from community notes. It emphasized that the list is not exhaustive and not “intended to constrain the Board from considering other factors which may be relevant”:
- Low levels of freedom of expression
- The absence of freedom of the press
- Government restrictions on the internet
- Low levels of digital literacy
- The ability, currently and in the past, to achieve the disagreement required for consensus [in the community notes algorithm]
The Board will consider these factors, among others.
The Board requests public comments that address:
- The risks and opportunities of crowd-sourced and community notes-style approaches to content moderation, particularly when it comes to potentially misleading content.
- The suitability and adaptability of consensus or bridging-based algorithms, which are employed in systems like community notes to identify and promote content that appeals across divided audiences, to different political contexts and information environments.
- Meta’s human rights responsibilities regarding the expansion and deprecation of products and programs, particularly those addressing misleading information.
- Challenges and best practices in risk assessment, monitoring, and mitigation for the global rollout of social media products, particularly in contexts of polarization, conflict or limited human rights protections.
- Research into the efficacy of responses to misleading information beyond content removal, such as fact-checking, labelling, reduced distribution, increased friction, and user-generated context. Additionally, research on avoiding bias in such responses.
- Studies that employ quantitative and/or qualitative research methods for identifying and measuring country-level factors that might impact the functioning of social media products across different contexts.
Prior decisions
The Board has engaged with Meta’s Misinformation policy, fact-checking and labelling of content in several cases, including:
- Alleged Audio Call to Right Elections in Iraqi Kurdistan, June 2025 (see Meta’s response here)
- Posts Supporting UK Riots, April 2025 (see Meta’s response here)
- Australian Electoral Commission Voting Rules, May 2024 (see Meta’s response here)
- Altered Video of President Biden, February 2024 (see Meta’s response here)
- Removal of COVID-19 Misinformation policy advisory opinion, April 2023 (see Meta’s response here)
- COVID Lockdowns in Brazil, August 2021 (see Meta’s response here)
टिप्पणियाँ
The comment was uploaded as a document attached in the public comments portal.
Assessing Meta's Plans to Expand Community Notes
Comment by the Jena Institute for Democracy and Civil Society
Jena (Germany), 8 December 2025
The idea behind Community Notes is that users should fact-check content to the best of their ability, thereby contributing to more effective moderation and fact-checking and reducing bias. Developed by Twitter in 2021, the program was scaled up and opened to all users in 2023. Since then, over 2.2 million Community Notes have been published in various languages on X.
Insights about the usage of Community Notes on X can contribute to a better understanding of the risks and opportunities of this tool in terms of more collaborative platform moderation, including on Meta platforms. Several studies have examined Community Notes on X, but few have addressed national peculiarities. In a recently published study by the Institute for Democracy and Civil Society, we identified 48,149 German-language Community Notes for subsequent analysis.
In the German-language context, we found that:
1. Only a few community notes are published as helpful (only 2,228 out of 35,508 German community notes — 6%). The ones that are published are released late, on average (median) 19 hours after publication.
2. Content that is polarizing is unlikely to receive a published Community Note. Participation among German-speaking users is low, and bipartisan consensus requires voters with different political backgrounds. This explains why 91% of all German Community Notes do not receive enough votes. In this context, potentially helpful community notes also remain unnoticed by the public.
3. Of a sample of German Community Notes, a majority (55.5%) are not concerned with fact-checking, but rather with opinion-making, mischief, and discussion threads about the quality of other Community Notes.
We will address each point in more detail below and discuss the mechanisms behind them.
*A) Publication rates and engagement decrease*
Looking at German-speaking Community Note takers, we can see that there are few incentives to engage in the long term. Of the 7,353 people who have written Community Notes, 3,175 have only written one. In contrast, 50% of all community notes have been drafted by only 7% of community note takers. This reflects an imbalance, showing that relatively few people participate in the program and that engagement rates are unevenly distributed.
Furthermore, there is a significant disparity between the number of requested and written Community Notes. Only 10% of requested community notes are written online, and fewer than 1% are published. This discrepancy underlines that the Community Note system does not encourage enough voluntary participation. This problem is exacerbated in smaller language communities.
Since March 2025, we have seen a decrease in participation in the Community Note program in Germany and globally. This can be attributed to the rise of verifying AI chatbots, such as Grok, which are now used much more often than Community Notes. The Community Notes program, as it is currently implemented on X, does not reward users who voluntarily moderate the platform with sufficient recognition. There is little public recognition, which motivates other collaborative knowledge production projects.
*Recommendation:* Before implementing Community Notes globally, Meta must implement measures to encourage user participation in the program.
*B Polarizing content and bridging-based algorithms*
The evaluation of a Community Note's ratings on X relies on a bridging-based algorithm that incorporates users' opposing "perspectives." Users are assigned to a particular "perspective" based on their previous rating decisions. This procedure ensures that only Community Notes that are widely agreed upon by large groups of users are displayed publicly.
While this mechanism primarily serves to safeguard the quality of Community Notes, it also makes it more difficult for a note to be displayed. In this sense, the algorithm operates as a filter for polarizing yet factually accurate content. Although the intention is the opposite, the publication mechanisms contribute to the acceptance of misleading information in polarizing posts. This can lead to further polarization.
Furthermore, community notes written in less widely used languages are at a disadvantage because they can only be rated by a comparatively small user base. Given the already low proportion of Community Notes rated as helpful, a more carefully calibrated adjustment of the algorithm's thresholds is necessary.
*Recommendation:* Meta should adjust the bridging-based algorithm according to political specifications in specific regions. Meta should furthermore prevent polarizing posts with misleading information from going unchallenged.
*C Community Notes and the efficacy of responses to misleading information*
Community Notes, as rolled out by X (and prospectively by Meta), are said to provide important context for misleading information, raise information accuracy, and push back on disinformation. However, many Community Notes do not truly add context to questionable posts.
In our study of German Community Notes, we found that most Community Notes do not aim to correct false claims or provide necessary information. In fact, we found that opinions, off-topic statements, and housekeeping notes outnumber rectifications based on facts.
Many of the notes are not intended for publication. Instead, they aim to discuss which Community Notes are appropriate and which are not. As neither X nor Meta provides a forum for this, the discussion takes place in the community note system. Thus, the number of Community Notes that adhere to the initial idea of collective fact-checking must be considered much smaller in future calculations.
*Recommendation:* Future rollouts of community notes should provide forums for community note takers to reduce the inappropriate use of community notes. Best practices for community noting are recommended, especially for regions where engagement needs to be fostered.
D Final remarks
For Community Notes to truly be a tool for less biased, bottom-up moderation, it needs a system of accountability for its participants. The success of this system depends on the long-term voluntary participation of many users, as well as features that recognize their contributions. If this cannot be guaranteed, the Community Notes system will quickly stagnate at an inadequate level. In any case, our study has shown that Community Notes cannot substitute for transparent content moderation.
The Institute for Democracy and Civil Society (IDZ) is a research institute dedicated to strengthening democratic culture online and locally. The IDZ combines social science and computational methods to study extremism, polarization, and conflict in the digital world.
The study is available here: https://www.doi.org/10.58668/matr/08.2.
As the Head of the Fact-Checking Unit at Annir in Libya, I am concerned about deploying a crowd-based moderation model like Community Notes in fragile information environments. In contexts marked by polarization and coordinated online behavior, such systems are highly vulnerable to manipulation. The “consensus” that the algorithm relies on can be artificially produced by organized groups, giving misleading narratives undue legitimacy.
Our work at Annir shows that Libya’s information ecosystem lacks the media literacy foundation needed for a crowdsourced model to function safely. Many misleading claims require deep contextual understanding, whether related to conflict dynamics, health misinformation, or cultural narratives, which volunteer contributors may not possess. In such contexts, trained fact-checkers remain essential to ensure accuracy and prevent harm.
For these reasons, discontinuing or reducing third-party fact-checking risks disproportionately harming Global South countries, where independent media systems are weaker and misinformation can escalate quickly. Third-party partners provide local expertise that global volunteer systems cannot replace.
I therefore recommend a hybrid approach: expanding Community Notes while maintaining and strengthening third-party fact-checking partnerships. Community Notes can support early detection and add user-generated context, but professional fact-checkers should continue to handle high-risk, sensitive, and context-heavy content. This combined model offers a safer, fairer, and more effective solution for diverse global contexts, including Libya.
Although community notes seem conceptually promising as a crowd-sourced solution to the perceived limitations of other forms of moderations, it remains too limited, and easily manipulated to serve as an effective standalone solution for combating misinformation. Evidence from both X and Meta shows extremely low publication rates, slow turnaround times, vulnerability to strategic misuse, and weak coverage of polarizing topics, all of which diminish the impact of community notes. Meta’s current implementation raises additional concerns: the program is small in scale, excludes advertising content, and lacks transparency around usage. These shortcomings create loopholes that are easy for bad actors to exploit.
Professional fact-checking on Meta’s platforms has demonstrated clear effectiveness in reduction of circulating misinformation through accurate labeling, downranking and reduced resharing, confirmed by user behaviour data. Instead of only considering in which countries not to roll out community notes - an approach that ignores ample empirical evidence - we recommend to integrate community notes with the established Third-Party Fact-Checking program, for example by providing fact-checkers with a “fast lane,” visibility into user-submitted notes, and mechanisms to accelerate high-quality contextual information. By highlighting the independence of fact-checkers and transparency of the community notes system, a blended approach leverages the strengths of both crowdsourced insight and professional expertise to reduce misinformation while protecting information integrity and the fundamental right to freedom of expression.
Please, find our detailed comment in the document attached.
The European Fact-Checking Standards Network (EFCSN) is the association and voice of independent European fact-checking organisations, representing 62 organisations from 36 countries.
FORUM-ASIA would like to submit the following comments for your consideration:
Factors to consider when adopting community-driven moderation, and how it could help or worsen misleading information
- Freedom of expression: Will the system expand the public’s ability to access contextual information, or will it disproportionately silence or bury views (especially minority or dissenting voices)?
- Non-discrimination and equality: Could socially marginalized groups be excluded from contributor pools or targeted by notes?
- Safety, harm and retaliation: Could contributors be exposed to state surveillance, reprisals, doxxing, or online/offline harm?
- Legal risks to contributors: Are there criminal laws (e.g., sedition, defamation, hate speech, terrorism) or broad emergency powers that could criminalize note writing or contextualization?
- Likelihood of state capture: Will governments try to manipulate contributor pools (e.g., coordinated volunteers, legal pressure on platforms)?
- Local language coverage: Could notes that don’t reflect local languages/cultural framings be incomplete, ineffective or misleading?
- Representation across viewpoints, regions, genders, ethnicities, and expertise: Is the contributor base broad enough to represent diverse viewpoints, or is it at risk of bias?
- Appeals and remediation: Can content authors challenge notes? Is there an independent appeals path for violations?
- Stop/rollback criteria: Are there clear thresholds (e.g., systemic harassment of contributors, sustained legal risk etc.) that justify pausing or reversing the launch if it is not working as intended?
The ability and limitations of consensus-based algorithms to function effectively in different political climates, especially where audiences are divided or information environments are weak
- Potential for self-censorship: In restrictive political climates, contributors may avoid participating out of fear, leading to misleading consensus and weakened correction capacity.
- Polarisation affecting consensus formation: In highly divided political climates, cross-perspective agreement becomes harder to reach, meaning fewer notes will be displayed and harmful content may go uncorrected.
- Risk of majority or dominant-group capture: In contexts where one political bloc or demographic dominates online spaces, “consensus” may simply reflect the views of the powerful, marginalizing minority voices.
- Susceptibility to coordinated manipulation: Consensus mechanisms can be gamed by organised political actors or disinformation networks able to flood the system with aligned contributors.
- Recognising that consensus is not the same as accuracy: Algorithms based on agreement may certify misleading or biased interpretations as “consensus” if coordinated networks or echo chambers reinforce one another.
- Reduced participation because of low civic trust: In weak information environments where mistrust of institutions is high, fewer users may participate as contributors, skewing the diversity and reliability of consensus signals.
- Limited effectiveness in contexts with poor media literacy: Where users struggle to distinguish reliable sources, consensus can normalize misconceptions rather than correct them.
Meta’s human rights responsibilities when expanding, scaling back, or ending products and programs that address misleading information
- Duty to prevent harm: Meta has a responsibility to assess and mitigate foreseeable risks to user rights - especially freedom of expression, safety, non-discrimination, and access to information - before expanding, reducing, or withdrawing tools that counter misleading content.
- Meaningful human rights due diligence: Any expansion or rollback must be guided by robust, ongoing human rights impact assessments that consider local political contexts, risks to vulnerable groups, and potential unintended consequences. These assessments should be independent and accessible to the public to ensure transparency.
- Non-discrimination and equity across regions: Meta must avoid creating unequal protections where users in some countries receive safeguards against misinformation while others - often in the Global Majority - are left exposed. Decisions to expand or limit coverage should not reinforce existing inequities in civic or information spaces.
- Transparency and access to information: When scaling back or ending programmes, Meta has a responsibility to provide clear, accessible explanations about why a product is being changed, what risks were assessed, and what alternative safeguards will remain. Sudden withdrawals undermine user trust and can worsen harmful information flows.
- Accountability and access to remedy: Users, human rights defenders, and affected communities should have channels to raise concerns, appeal adverse impacts, and receive timely responses. Ending or cutting back a product does not remove Meta’s obligation to remedy harm if misinformation spikes or if communities are disproportionately affected.
- Protection of contributors: If changes to a product increase risks to those who generate or rely on contextual information (e.g., Community Notes contributors, human rights defenders, journalists etc.), Meta must implement additional safeguards such as anonymity and emergency protocols, or reconsider the decision entirely.
The effectiveness of interventions beyond content removal, such as fact-checking, labelling, reduced distribution, added friction, and user-generated context, and how to avoid bias in these approaches
- Fact-checking may enhance accuracy but depends on local capacity: Independent fact-checking can be a good way to correct information without censoring speech, but its effectiveness is limited if fact-checkers are biased, lack linguistic reach, face political pressure, fear reprisals, or cannot keep pace with high-volume misinformation.
- Labelling and contextual warnings may help users make informed choices if bias is accounted for: Clear, well-designed labels (e.g., “missing context,” “disputed claim” etc.) can significantly reduce belief in false content, especially when they explain why the content is misleading. However, labels must be applied consistently to avoid bias.
Country-level factors that influence the functioning of social media products and how these factors can be studied or measured
- Strength of civic space and human rights protections: Restrictions on expression, press freedom, or civil society influence user behaviour and the safety of contributors.
- Risk of state interference, surveillance, and censorship: High surveillance or authoritarian contexts increase risks for users and contributors and can further distort platform behaviour.
- Political stability and level of polarisation: Highly polarised or unstable political environments affect how users interpret information and how easily products can be weaponised.
- Media literacy and information ecosystem health: Countries with weak independent media or low digital literacy are more vulnerable to mis/disinformation and manipulation.
- Prevalence of online harassment and coordinated abuse: High levels of trolling, gendered harassment, or coordinated political attacks reduce participation in community-driven features and distort consensus.
We hope that these questions and suggestions will be taken into consideration when determining the expansion of the Community Notes programme.
Community Notes as One Layer in a Multi-Stack Moderation System and Why Fact-Checking Remains Critical for India’s Information Integrity
Argument Summary
Community Notes (CN) is an addition to Meta’s integrity toolbox, offering community participation, and contextual explanations around disputed content. However, CN’s crowdsourced design also carries structural limitations that make it insufficient as a standalone solution in India’s diverse and high-stakes information environment. This is also true for many other countries and regions in the world, and especially crucial for India as it is the largest market for Meta's products.
As smartphone penetration deepens and data costs remain low, a new wave of users from tier-2, tier-3 cities and rural areas are accessing content in Hindi, Tamil, Telugu, Bengali, Marathi, and other local languages. With 22 officially recognized languages under the Indian Constitution’s Eighth Schedule and 121 languages with significant user bases, India is home to an extraordinary linguistic diversity. While this linguistic richness has historically coexisted with a public sphere often dominated by English and Hindi, particularly in traditional media and early internet phases, with the next wave of internet users emerging from non-metro regions, language is no longer a barrier to participation. Instead, it is the medium of choice.
In addition to this, take into account India’s low digital literacy with only around 38% of the households reported to be digitally literate. This makes the majority of the non-English speaking new internet users more vulnerable to misinformation.
Our analysis of Community Notes on X’s platform, shows that since its launch, only about 2,200 notes have been published in Indian languages, written by only 805 unique contributors, as of December 2025. Of these, just 55 notes in Indian languages were rated as “currently helpful”, as of December 6, 2025. (Refer data in Table 1 and 2 below)
Independent studies examining Meta’s Community Notes reflect similar patterns of severe linguistic imbalance. Because of high consensus thresholds, notes in Indian languages rarely get published, leaving large regional-language spaces underserved.
Moreover, studies tell us that CN’s operational design is well-suited for correcting simple factual inaccuracies but is not equipped to evaluate complex or harm-based narratives such as gendered, communal, or caste-driven misinformation where trained experts are essential.
Moreover, CN’s reliance on cross-group agreement significantly delays visibility (averaging 14 days globally), while notes that reference professional fact-checkers appear earlier, are more trusted, and achieve higher visibility. For these reasons, Community Notes should function as one layer within a multi-stack approach, complemented, not replacing third-party fact-checking. This analysis draws on independent academic and institutional studies of X’s Community Notes model, which informs Meta’s CN architecture, given the limited India-specific transparency data available from Meta platforms.
Key Arguments Against Replacing Third-Party Fact-Checking With Community Notes Model In India
South Asian languages are severely underrepresented, creating an information integrity gap. South Asian languages (Hindi, Urdu, Bengali, Tamil, etc.) account for only 0.094% of the global CN corpus, leaving vast vernacular online spaces vulnerable to unchecked misinformation. The CN interface, guidelines, and sign-up process are still heavily English-centric. (CSOH, 2025).
The TIA also extracted data on X’s Community Notes to validate the share of Indian languages. As per our data, more than 25 lakh notes were written since the inception of Community Notes till December 5, 2025 and only 0.894% were written in Indian languages. Only 55 notes were rated as ‘helpful’.
Table 1: Share Of Notes In Indian languages In Community Notes Published on X In India, since inception
Labels
Currently rated helpful
Currently not rated helpful
Needs more ratings
Blank
Grand Total
Share in total CN
Bengali
9
1
10
0.0004%
Gujarati
1
7
2
10
0.0004%
Hindi
40
11
1063
150
1264
0.0504%
Kannada
8
5
13
0.0005%
Malayalam
58
25
83
0.0033%
Marathi
58
12
70
0.0028%
Punjabi
2
2
0.0001%
Tamil
4
2
193
25
224
0.0089%
Telugu
1
1
18
1
21
0.0008%
Urdu
9
3
514
18
544
0.0217%
Grand Total
55
17
1928
241
2241
0.0894%
The consensus algorithm fails in low-volume language communities. Notes in South Asian languages face structural barriers leading to a publication rate of only 2.30% compared to 8.25% globally. This percentage is lower for Indian languages, as shown in Table 1 above. High-quality notes often remain "stuck in limbo" because low reviewer density prevents the bridging algorithm from finding the necessary cross-cluster agreement. This requires vernacular notes to clear a substantially higher threshold of review (e.g., typically needing over 80 ratings to publish). (CSOH, 2025)
Increased participation does not guarantee success, leading to bottlenecks. The system fails to scale efficiently; as seen globally, even where contributor volume increases, publication success can be low (Spanish notes still lag) or cause new bottlenecks (too many English notes for raters to choose from), resulting in fewer notes breaking through. This undermines the promise of collective moderation. (DIA 2025)
Table 2: Number of Unique Contributors In CNs In Indian Languages Since Inception Till Dec 6, 2025
Language
Unique Contributors
Share In Total
Bengali
8
0.002%
Gujarati
3
0.001%
Hindi
389
0.107%
Kannada
10
0.003%
Malayalam
26
0.007%
Marathi
45
0.012%
Punjabi
2
0.001%
Tamil
100
0.028%
Telugu
18
0.005%
Urdu
204
0.056%
Grand Total
805
0.222%
CN is far too slow to counter viral harms, particularly during crises. Despite improvements, the average time for a note to go public is still 14 days. This is insufficient to counter online harms that spread within hours (DIA 2025). Furthermore, CN usage in India surges only reactively during high-stakes events like the General Election, meaning hundreds of drafts languish during the weeks voters need them. (CSOH, 2025)
The system is vulnerable to partisan misuse in polarized environments. In South Asia, a significant portion of notes carry political biases (46.0%) or potentially harmful/divisive tone (4.5%). Examples include notes using personal slurs ("Italian-mafia supporters") or political rhetoric rather than fact-checking. This creates the risk of "crowdsourced propaganda," leading users to distrust the feature as merely another arena for political mud-slinging (CSOH, 2025).
Relevance of Fact-Checks
Expert fact-checking provides essential speed and quality indicators. Community notes that cite evidence from fact-checking organizations (the third most used global reference) are significantly more likely to become visible (12% visibility rate vs. 8.3% overall). Crucially, Community Notes that cite a fact-checker are considered more useful by users and appear on tweets more quickly—90 minutes earlier than general notes (Maldita 2025).
Expertise is crucial for nuance and local context. Experts, such as journalists and fact-checkers, are "deeply attuned to local languages, cultures and socio-political nuances," and are important for assessing the "check-worthiness" of narratives and their potential harm. This specialized knowledge is precisely what the CN model, which is skewed towards Western epistemology, lacks (Adriana, Nadia 2025).
CN's consensus requirement conceals useful expert-backed information. X's focus on "consensus among users who usually disagree" rather than factuality is a "false premise of equaling truth with consensus". As a result, over 85% of high-quality notes proposed that cite independent fact-checking organizations are still not visible (Maldita 2025).
The two approaches can be made complementary, instead of being contradictory. The work of professional fact-checkers and community involvement are "essential and complementary" (Maldita 2025). Meta should avoid repeating the mistakes of X and should collaborate with professional fact-checking organizations. Solutions should involve a close collaboration between experts, systems, and non-experts to maintain quality and accountability.
Conclusion
Community Notes can play an important role in India, but it cannot, by design, address the country’s linguistic breadth, polarized discourse, rapid misinformation spread, or harm-based narratives on its own. This is why third-party fact-checking must coexist as an additional layer, especially since evidence suggests that CN performs better when informed by expert fact-checkers.
Examples of Community Notes
Automated content moderation systems can miss out on context and nuance
The Tweet below is written in Hindi language by the president of the Delhi unit of Aam Aadmi Party (AAP), a political party in India.
The tweet says, “11.11.2025 Chief Minister (CM) Rekha Gupta met the patient first. The next day on 12.11.2025, the Prime Minister (PM) met the same patient. New costume. New Plaster.”
The tweet shares two images in which CM Rekha Gupta is seen interacting with a patient in a hospital on the left and the image on the right showing PM Narendra Modi meeting a patient on another day. The tweet was published on November 11, 2025, a day after a car exploded near Red Fort in Delhi killing several civilians. The views on the tweet shows its virality.
While the tweet does not make any direct claims, it insinuates that when the Prime Minister visited the patient he got a plaster while when the Chief Minister had visited he did not have a plaster, alluding that he is faking his injuries. In the comments section, we see that several users have posted a timeline of the Delhi CM meeting the patient several hours after the blast and the PM meeting the patient a day later explaining that the state of the patient, including his clothes, his appearance etc. are bound to be different and it cannot be used as evidence that the patient is faking his injuries. The tweet has no community note till date but there are fact-checks on the internet that have fact-checked the post verifying the timeline of events.
This tweet is an example of why context and nuance matter in combating misleading narratives that do not include a direct claim. The reason why such tweets are able to spread disinformation is because there have been instances in the past where images or associated text featuring political personalities meeting civilians in a crisis etc. have been debunked as manipulated. This is the kind of nuance and context that fact-checkers understand because they are part of a system that builds institutional knowledge.
Polarising tweet targeting a demographic doesn’t feature CNs as notes remain stuck waiting for cross-idealogical ratings
The claim made in the tweet below has been debunked by a fact-check article which has also been submitted in the note by a contributor, however, the note remains unpublished. This shows how cross-idealogical voting does not increase or ensure neutrality in information sharing on online spaces.
A tweet in Marathi consisted of an AI-manipulated video that had notes but didn’t pass the rating threshold.
End Of Report
Meta’s plan to use X’s expansion of community notes as a precedent needs to take inherent concerns of that action into account.
I research the extent of exposure journalists in the Asia Pacific face to technology facilitated hazards and impact.
While X moved ahead with promoting community notes and dialling down inhouse human moderation, we witnessed journalists and other information specialists in the Asia Pacific either move out of X or decrease engagement to just being present in the platform.
In August 2023, Australia’s public broadcaster ABC removed dozens of accounts linked to it on X as means to limit exposure to hazards. Internal ABC sources have said that the number of accounts deleted could be as high as 70. They were wilted down to four. Among those who deleted their accounts on X was one of ABC’s most prominent journalists, Lisa Millar, who hosted the flagship Morning Show. In my work I have spoken with dozens of journalists spread across the region who either have removed themselves from the platform or resorted to silence since Twitter changed over to X.
While X’s emphasis on community notes and easing of human moderation were not the only reasons for such exits, they played an influential role. Journalists we have spoken to relate how the lack of moderation increases toxicity, abuse and vitriol directed at them.
This trend is compounded by the increase in disinformation and misinformation which gets ramped up when moderators are removed. What we hear often is that X abdicated its duty of care for its community and showed disregard for the potential for harm, including psychosocial harm, which content on the platform could potentially prompt.
When information specialists like journalists resort to digital silence, the community is denied verified and authenticated information disseminated by a professional trained and experienced in the craft. That information void is then taken over by disinformation and misinformation which can be gamed and scaled to completely overwhelm the information architecture. We have seen this happen in Asia and the Pacific at local, national and regional scale.
The likelihood of this repeating and at a much wider scale on Facebook is very real if Meta does not invest in attempts to ensure that digital hate and toxicity aimed at journalists in this case, are mapped, monitored and prevented at the very outset and not when they have proliferated the space.
Our ongoing work at the Centre for Journalism and Trauma Asia Pacific and my own research at the Adelaide University has provided data to indicate that journalist practice digital silence when exposure levels are overwhelming them.
It is not only journalists who face this threat, anybody who is on any digital platform is at risk. That risk is increased if social media companies like Meta fail to acknowledge it and act on it.
Submission to Meta Oversight Board: Policy Advisory Opinion on Community Notes Expansion
Submitted by: Dr Sanjana Hattotuwa
Date: 7 December 2025
This submission addresses the Board's specific questions whilst examining Meta's broader failure to conduct human rights due diligence, the technical limitations that make community notes unsuitable as fact-checking replacements, and the asymmetric harms Global South communities will face from this transition.
High-level observations
Meta's request asks which countries to exclude from community notes rather than demonstrating why any country should be included. This inverts human rights due diligence. The burden of proof lies with the company to show expansion will not cause harm, not with affected communities to prove they deserve protection.
The Oversight Board faces a choice: provide cover for Meta's predetermined expansion or demand the company meet its human rights obligations before proceeding. The evidence supports only one conclusion: Meta should restore fact-checking partnerships, conduct comprehensive impact assessments with meaningful community consultation, and suspend community notes expansion until independent evaluation confirms the approach functions without enabling the harms the company's platforms have repeatedly caused. The question is not how Meta should expand community notes but whether expansion should occur at all given the company's failure to establish basic prerequisites: demonstrated effectiveness, appropriate safeguards, adequate language capacity, and credible commitment to human rights principles.
Community notes could provide value when integrated alongside, rather than replacing, professional fact-checking. Effective implementation would require Meta to restore fact-checking partnerships, use professional assessments to reduce distribution of false content, and allow community notes to provide additional context on borderline cases. This layered approach addresses different aspects of information disorder: professional fact-checkers establish ground truth, algorithmic interventions reduce viral spread, and community contributions surface local context. The current framing eliminates this potential. Meta presents community notes as cost-saving measures replacing labour-intensive partnerships rather than as complementary tools enhancing platform integrity. Internal research Meta commissioned in 2023 found fact-checking partnerships provided substantially more value per dollar spent than algorithmic interventions alone, findings the company has not publicly disclosed.
Meta's human rights responsibilities regarding product expansion and deprecation
The 2021 Corporate Human Rights Policy established binding commitments
Meta's 2021 Corporate Human Rights Policy made explicit pledges the company is now violating. The policy committed to paying “particular attention to the rights and needs of users from groups or populations that may be at heightened risk of becoming vulnerable or marginalised” and to “identifying relevant such groups for each context, undertaking meaningful engagement to hear their hopes and concerns.” The January 2025 policy changes were announced without consulting civil rights advisory committees, the Oversight Board, or affected communities. The NAACP Legal Defense Fund withdrew from Meta's civil rights advisory committee, stating: “Meta made these content moderation policy changes without consulting or warning us, and many of the changes directly conflict with guidance from LDF and partners.”
On human rights defenders specifically, the 2021 policy stated: “We condemn all threats, acts of intimidation and retaliation, persecution, and physical and legal attacks against human rights defenders... We strive to support their important work.” The revised Hateful Conduct policies now explicitly permit calling LGBTQ+ people “mentally ill” and allow referring to immigrants as “filthy,” directly enabling attacks against categories of human rights defenders Meta pledged to protect. The 2021 policy committed to human rights impact assessments: “When faced with projects or decisions that may significantly affect human rights, we will undertake a human rights impact assessment.” Meta has not disclosed conducting such assessments before eliminating fact-checking or revising hate speech policies. The Oversight Board's April 2025 recommendations noted policies were “announced hastily, in a departure from regular procedure, with no public information shared as to what, if any, prior human rights due diligence the company performed.”
United Nations Guiding Principles require ongoing due diligence
The UN Guiding Principles on Business and Human Rights, which Meta's 2021 policy claims to follow, require companies to “carry out human rights due diligence” including assessing “actual and potential human rights impacts.” Principle 18 specifies this process should “involve meaningful consultation with potentially affected groups and other relevant stakeholders.” Meta's community notes expansion proceeds without disclosed due diligence. The company describes possessing “limited data from the US beta rollout” whilst requesting guidance on global expansion, an admission that impact assessment has not occurred. Principle 19 requires prevention and mitigation measures proportionate to severity of impacts. Meta's elimination of fact-checking before establishing whether community notes function at scale inverts this obligation.
The request asks the Board to identify factors for “deciding which countries to omit from community notes” rather than which contexts might safely accommodate them. This framing assumes expansion as default, with exclusions requiring justification. Human rights due diligence requires the opposite: companies must demonstrate expansions will not cause harm before proceeding. Principle 20 requires companies to “track the effectiveness of their response” through “appropriate qualitative and quantitative indicators” and “draw on feedback from both internal and external sources.” Meta eliminated fact-checking in January 2025, began testing community notes in March 2025, and now solicits guidance on global expansion in December 2025, insufficient time to assess effectiveness or gather external feedback on outcomes.
Challenges in risk assessment for global rollout in polarised contexts
Meta's proposed factors are insufficient and incorrectly weighted
Meta's list of factors for potential country exclusions includes “low levels of freedom of expression,” “absence of freedom of the press,” “government restrictions on the internet,” “low levels of digital literacy,” and “ability to achieve the disagreement required for consensus.” These factors misunderstand how community notes fail in precisely these contexts. Countries with low press freedom are where professional fact-checking is most critical. When independent media cannot operate, crowd-sourced moderation defaults to state-aligned narratives. Meta's factor list treats absence of press freedom as reason to exclude community notes rather than reason to maintain professional fact-checking. This inverts the human rights analysis: vulnerable contexts require more protection, not less. Similarly, “government restrictions on the internet” correlate with state manipulation of online discourse. Ethiopia's government implemented internet shutdowns during the Tigray conflict whilst operating sophisticated propaganda networks on Facebook. Community notes in such contexts would amplify rather than counter state disinformation because contributor pools skew toward regime supporters, those with continued access, approved speech patterns and incentive structures favouring alignment. “Low levels of digital literacy” appears as potential exclusion criterion whilst research shows these populations face asymmetric vulnerability to sophisticated disinformation. The logical response is strengthened professional fact-checking, not reliance on crowd-sourced contributions from populations lacking tools to assess source credibility. The factor “ability to achieve the disagreement required for consensus” acknowledges the algorithm's fundamental flaw: it requires polarisation to function. In authoritarian contexts with suppressed opposition or societies with cross-partisan agreement on targeting minorities, the algorithm fails precisely because it cannot identify “disagreement.” Yet Meta frames this as reason to exclude countries from community notes rather than reason to question whether the algorithm itself is appropriate.
Language capacity represents a binding constraint
Meta's factor list omits the most determinative element: language. Community notes require sufficient contributor density in each language to generate consensus ratings. Languages with millions of speakers but insufficient Meta user populations, Sinhala, Burmese, Amharic, Tigrinya, will lack contributor bases regardless of other factors. Languages Meta classifies as “low-resource” face systematic disadvantage: fewer moderators, less sophisticated algorithms, higher error rates and slower response times. Code-switching and multilingual content defeat both automated systems and community notes. Indian social media frequently combines Hindi, English and regional languages within single posts. Ethiopian content mixes Amharic, Oromo and English. Somali diaspora communities write in Somali using Latin rather than Arabic script, rendering automated translation systems ineffective. Community notes require contributors who understand all languages within a post, a rare combination.
Conflict contexts require the opposite approach Meta proposes
Meta asks how to assess risks in “contexts of polarisation, conflict or limited human rights protections” whilst pursuing policies that exacerbate precisely these conditions. Research on platform harms during conflict shows consistent patterns: coordinated networks manipulate engagement metrics, state actors suppress opposition voices, propaganda overwhelms accurate information, and violence follows dehumanising rhetoric. The pattern repeats: civil society warns Meta about escalating hate speech, the company responds inadequately due to language constraints, violence occurs, Meta apologises and promises improvement, then deprioritises the same markets when conflicts subside. Community notes could worsen this cycle by eliminating the professional partnerships that provided Meta's only reliable early warning in conflict contexts. Trusted partner networks in conflict zones include organisations with on-the-ground presence, relationships with affected communities, and expertise in local dynamics. These partners contextualise content that appears innocuous to external reviewers but telegraphs violence to intended audiences. Community notes rely on users who may themselves be conflict participants, lack protective protocols for sensitive cases, and have no accountability beyond Meta's opaque contributor scoring.
Recommendations to the Oversight Board
1. The Oversight Board should recommend Meta suspend community notes expansion until the company conducts and publicly discloses comprehensive human rights impact assessments for each proposed market. These assessments must include meaningful consultation with affected communities, independent evaluation of moderation capacity in local languages, and transparent reporting on professional fact-checking partnerships Meta eliminated.
2. The Board should establish minimum thresholds Meta must meet before expanding community notes to any market: sufficient contributor density in local languages, demonstrated ability to generate consensus within timescales that prevent viral spread, independent validation that notes address substantial percentages of misleading content, and evidence that consensus mechanisms do not systematically disadvantage minority perspectives.
3. Meta must commit to elevated moderation capacity in conflict contexts rather than reduced oversight. Countries experiencing active armed conflict, recent mass atrocities, or patterns of ethnic violence should receive enhanced professional fact-checking, increased human review and expedited escalation processes, the opposite of what community notes provide. The company should establish clear triggers for suspending community notes in contexts where coordinated manipulation or state capture becomes evident.
4. The Board should require Meta disclose internal research on fact-checking effectiveness, community notes performance data disaggregated by country and language, algorithmic audit results for consensus mechanisms, and comparative assessments of intervention effectiveness across markets. This disclosure should occur before any expansion decisions, subject to independent verification by qualified researchers, and updated quarterly to enable ongoing assessment.
5. Meta's 2021 Corporate Human Rights Policy should be enforced as binding commitment. The company must demonstrate how community notes expansion complies with pledges to protect vulnerable populations, support human rights defenders, conduct impact assessments and implement oversight mechanisms. Where compliance cannot be demonstrated, expansion should not proceed.
6. The framing of this request reveals Meta's approach: the company eliminated proven safeguards, imposed changes contradicting its human rights commitments, ignored warnings from civil society organisations, bypassed the Oversight Board's consultation, and now seeks validation for predetermined expansion. The Board's response should centre accountability rather than enablement.
Case: Assessing Meta’s Plans to Expand Community Notes
Date: December 3, 2025
1. Presentation of the organization and purpose of the comment
Cubalex is a non-profit organization dedicated to the defense and promotion of human rights in Cuba. Founded in Havana in 2010 as an independent legal aid center, it was forced into exile in 2017 and currently operates from the United States. From abroad, it provides free legal assistance, documents human rights violations, and represents cases before the Inter-American Commission on Human Rights (IACHR), UN mechanisms, and national courts. Its work has been internationally recognized through reports submitted to the Committee on the Elimination of All Forms of Discrimination against Women (CEDAW), the Committee against Torture (CAT), the Committee on the Rights of the Child (CRC), and in the Universal Periodic Reviews of 2013, 2018, and 2023.
This commentary responds to the Oversight Board's request to assess the global expansion of Community Notes. It explains how Meta's product decisions can affect access to information and justice in countries with limited civic space. It identifies two main risks. The first is the impact of geographic restrictions. The second is the impact of sanctions on compliance practices. Both of these harm users living in high-risk contexts. They also affect independent organizations that use Meta's platforms to inform the public, document abuses, and create safe channels for public participation. The document proposes human rights safeguards to prevent patterns of exclusion already observed in other Meta products. It uses the WhatsApp Business restriction in Cuba as a concrete example. It shows how poorly calibrated global decisions can harm vulnerable societies and should be taken into account by the Oversight Board.
2. The Cuban context: why it is a high-risk country for products like Community Notes
Meta has requested guidance from the Oversight Board on the factors to consider when deciding whether to implement Community Notes in a given country. These factors include limited freedom of expression, the absence of an independent press, government restrictions on internet access, low digital literacy, contexts of state surveillance, and difficulties in generating authentic dissent within the consensus algorithm. Cuba brings together all these elements simultaneously and seriously, making it an emblematic case of high risk for the expansion of this type of tool based on citizen participation.
The Cuban press system is not free. Independent media and critical voices face criminal and administrative persecution, constant surveillance, police summonses, and smear campaigns. This repression makes it impossible for a pluralistic media ecosystem to exist where alternative narratives or independent fact-checking can circulate. At the same time, internet access is strictly controlled by the state through a telecommunications monopoly. It is expensive, unreliable, and subject to selective platform blockages during protests or other sensitive events, severely limiting the population's ability to stay informed and communicate.
These structural information barriers are compounded by low digital literacy and a heavy reliance on a few apps (mainly WhatsApp, Facebook, and Instagram) that serve as the primary source of information for most citizens. The lack of knowledge regarding verification, digital security, and traceability increases their vulnerability to product changes or restrictive policies implemented by Meta.
Finally, the climate of fear and widespread surveillance creates significant risks of participatory tools being captured by state or paramilitary actors. In an environment where public dissent leads to real reprisals, participation in open systems like Community Notes can be distorted, favoring official narratives and limiting the diversity of opinions. Taken together, these conditions make Cuba an extreme case of how structural factors can profoundly affect the functioning of digital products based on social consensus and voluntary participation.
3. The WhatsApp Business case: an example of overcompliance and structural damage
As part of its remote legal assistance work, Cubalex attempted to integrate the WhatsApp Business API to professionalize services for victims and their families in Cuba, improve response times, and ensure more efficient and secure information management. Upon initiating the technical process, the organization encountered an explicit Meta restriction that prevents individuals with Cuban phone numbers (+53) from receiving messages from WhatsApp Business accounts. This limitation applies even when the sending organization (in this case, Cubalex) is located outside of Cuba, does not conduct business in Cuba, and uses the tool exclusively for humanitarian purposes. Between January 14 and 17, 2024, Cubalex engaged in extensive communication with the WhatsApp Business support team, providing technical evidence and explaining in detail the humanitarian impact of the measure; however, Meta confirmed that Cuba was included in its geographical restrictions, meaning it was banned from receiving messages on the platform.
The legal analysis conducted by Cubalex regarding U.S. export control regulations (eCFR – 15 CFR Parts 730 and 746) demonstrates that the embargo against Cuba primarily restricts the export of goods, financial services, and dual-use technologies, but does not prohibit free communication services such as messaging, email, or social media. These tools are, in fact, often covered by general licenses and information exceptions. Therefore, there is no legal provision that compels Meta to prevent Cuban users from receiving messages from the WhatsApp Business API. The restriction appears to stem more from a corporate policy of overcompliance, aimed at minimizing regulatory risks through indiscriminate blocking, even when this affects humanitarian activities and fundamental rights.
The consequences of this decision are profound. The measure disrupts the primary communication channel used by victims and their families, prevents the automation of essential processes for handling urgent cases, and forces them to resort to less secure channels that are more vulnerable to state surveillance, thereby increasing personal risks and reducing the ability to document violations promptly. It also frustrates Cubalex's organizational development efforts, as the organization had invested resources in adopting a key tool to improve its efficiency and data protection. This case demonstrates how product decisions made under a maximalist legal-compliance logic can cause structural damage to access to justice and to the protection of vulnerable populations in authoritarian contexts.
4. Relevance of the case for the expansion of Community Notes
The Oversight Board has requested input on the risks and opportunities of moderation models based on community-generated content, the adaptation of consensus algorithms to different political contexts, Meta's human rights responsibilities in expanding or withdrawing products, and the challenges of assessing country risks. In this context, Cuba's experience with the blocking of the WhatsApp Business API offers particularly relevant lessons for the advisory opinion on Community Notes.
The first lesson relates to the risk of preemptively excluding countries without conducting a differentiated impact analysis. The factors Meta identified as risk signals for Community Notes expansion. These scenarios fully reflect the Cuban situation. If Meta applies the same logic used for WhatsApp Business, Cuba could be automatically excluded from the program without a specific assessment of the consequences for the population. It's also possible that the product will be formally implemented but prove inoperative due to easy capture by state actors. In both scenarios, the dynamic of excessive enforcement would be repeated, restricting rights without providing additional protection and leaving those already living in an environment of censorship and repression without avenues for participation.
The second lesson points to the opposite risk: deploying Community Notes in repressive countries without adequate safeguards. In Cuba, where visible dissent carries significant personal risks, users who contribute critical notes can be identified and face reprisals. At the same time, self-censorship and coordinated action by pro-government accounts can distort the consensus patterns that feed the algorithm, creating “contexts” that legitimize official narratives and silence independent documentation of abuses. Under these conditions, a tool designed to mitigate disinformation could become a mechanism that reinforces state propaganda.
Finally, how Meta and the Oversight Board address the relationship between country risk, human rights obligations, and algorithmic design will have implications that extend far beyond Community Notes. Decisions made in this process will influence policies on geographic restrictions, access to advanced tools for independent organizations, and the development of humanitarian exception mechanisms. Therefore, it is essential that the advisory opinion explicitly address the problem of corporate overcompliance and its impact on civil society in sanctioned countries.
5. Recommendations to the Oversight Board
In light of the risks identified in the Cuban case, the Oversight Board's advisory opinion must include recommendations to address the structural gaps Meta presents when evaluating high-risk countries. The company must avoid automatic exclusion approaches based solely on sanctions status or general indicators of freedom of expression, as these categories do not reflect the true complexity of authoritarian contexts or their effects on civil society. Before excluding or limiting a country's access to Community Notes, Meta should be required to conduct a rigorous Human Rights Impact Assessment (HRIA) capable of identifying the foreseeable effects on human rights defenders, journalists, victims, and independent organizations, as well as the existence—or absence—of alternative safe channels. Only based on an individualized analysis can disproportionate decisions be avoided that, under the guise of regulatory caution, generate greater information isolation.
In parallel, Meta must develop formal humanitarian or civil society exception pathways for restricted products in sanctioned contexts. Such exception pathways must include transparent and accessible mechanisms for organizations like Cubalex, which operate from outside the country, do not maintain commercial relations with Cuba, and rely on digital tools to provide legal assistance and document human rights violations. The absence of such mechanisms is precisely what has allowed the restriction on the WhatsApp Business API to cause such severe and unjustifiable harm.
Similarly, in countries where dissent carries personal risks, Meta cannot assume the existence of genuine and spontaneous “consensus.” Consensus must be understood as a hypothesis requiring empirical verification, especially when self-censorship, state surveillance, and coordinated action by pro-government actors can distort both contributions and the very dynamics of Community Notes. Therefore, strengthened safeguards are needed, including robust anonymization systems, protection mechanisms for those making critical contributions, and independent audits capable of detecting algorithmic manipulation or biases induced by repression. Without these safeguards, the tool risks amplifying state narratives rather than balancing them.
Another essential aspect is consistency across product decisions. Meta's approach to Community Notes should also be reflected in its approach to other tools, such as the WhatsApp Business API. A consistent policy requires reviewing geographic restrictions that disproportionately affect humanitarian and human rights organizations. Furthermore, the company should maintain an ongoing dialogue with civil society organizations in high-risk countries to identify unforeseen effects, design reasonable exceptions, and monitor long-term impacts.
The Oversight Board has already underscored the need for such an approach in previous decisions. In the case of "Call for Women's Protest in Cuba," the Board reversed the removal of a protest video and strongly criticized Meta for failing to adequately consider the country's political and repressive context when applying its policies. The Board emphasized that freedom of political expression in Cuba requires greater protection, given state persecution of dissent and severe limitations on internet access. It also concluded that Meta over-applied its policies by classifying as hate speech expressions that were, in fact, part of a legitimate call to protest, highlighting the need for trained personnel with regional knowledge and more robust contextual analysis mechanisms.
These conclusions fully align with Cubalex's observations regarding Meta's WhatsApp Business restrictions, in which Meta adopted preventative measures without proportionate analysis, directly impacting victims and human rights defenders. Therefore, based on lessons learned from that case, we propose that the advisory opinion recommend the creation of a Civil Society Context Advisory Panel composed of independent organizations with experience documenting human rights in high-risk contexts. This panel would allow Meta to receive non-sensitive contextual information that would improve its understanding of repressive patterns, assist in identifying risks of over-enforcement, support the development of humanitarian exceptions, and ensure that product decisions comply with both human rights standards and current U.S. regulations. The existence of this mechanism would help implement the Board's repeated recommendations, strengthen contextual analysis, prevent unintended harm resulting from over-compliance, and ensure that Meta's products do not inadvertently become tools that exacerbate information isolation and the vulnerability of civil society in authoritarian countries.
6. Conclusion
The case of Cuba demonstrates that product decisions made by Meta (when designed with a maximum regulatory risk approach, lacking sufficient contextual analysis or exception mechanisms) can leave entire communities without access to essential communication, documentation, and advocacy tools. By operating in this way, the company not only unintentionally limits civil society's ability to protect rights and denounce abuses but also effectively reinforces the objectives of authoritarian regimes that benefit from information isolation and the absence of secure channels for citizen participation.
The expansion of Community Notes presents an opportunity to reverse this trajectory and establish stronger protection standards in high-risk contexts. Suppose the Oversight Board incorporates in its advisory opinion an explicit critique of overcompliance, a requirement for human rights impact assessments, and an obligation for Meta to create precise exception mechanisms for humanitarian and civil society actors. In that case, it will not only help make Community Notes a fairer, safer, and more effective tool but also ensure they are used in a fair, safe, and effective way. Still, it will also send a powerful normative message that will influence Meta's entire product ecosystem. Robust guidance from the Board could transform how the company assesses country risks, designs algorithms, and responds to repressive environments.
Cubalex remains available to the Oversight Board to provide further information and additional evidence, and to participate in future technical consultations related to the human rights situation and the specific characteristics of the Cuban digital ecosystem. We are committed to collaborating in any process that helps ensure that Meta's product decisions are made responsibly, with contextual sensitivity, and in accordance with international human rights standards.
The very design of the consensus algorithm makes Community Notes unsuitable to moderate entire categories of misleading content. By definition it is almost impossible to find consensus about hotly contested partisan issues. So whatever the facts may be, the consensus algorithm all but guarantees that most community notes in such debates will never see the light of day.
In effect, the algorithm powering Community Notes allows partisans to censor the facts in notes they don't like. This is an excellent feature if the goal is not to offend or irritate partisan users, but it is not so great if the goal is to ensure information integrity by showing factual information under posts that contain false information.
Information integrity should not be about "appealing" to as many audiences in "different political contexts" as possible but about providing correct and verified information to *everybody*. Community Notes approach to be "appealing" consists of selectively suppressing the notes that won't appeal to partisans.
Also see: "Community Notes are Vulnerable to Rater Bias and Manipulation", https://arxiv.org/html/2511.02615v1
Even though the voting histories of Community Notes users are shielded by pseudonyms there also is a serious risk of data abuse inherent in the way the consensus algorithm operates. It needs to keep track and process information about how every user voted on sometimes highly politically charged notes in order to be able to see which users have historically different voting histories so this information can be fed into the consensus algorithm. This information could be used to build detailed profiles about the political, religious and philosophical views of these users. Were this information to be leaked, abused or subpoenaed by governments, the consequences could be catastrophic. It would make the Cambridge Analytica scandal look small by comparison. In the European Union, the GDPR literally prohibits collecting and processing such data outside a very narrowly defined scope of exceptions.
This is not hypothetical: researchers have been able to infer political leanings from Community Notes voting history: "Algorithmic resolution of crowd-sourced moderation on X in
polarized settings across countries", https://arxiv.org/pdf/2506.15168
Compared to a third-party fact checking system, the current implementation of Community Notes by Meta in the U.S. is significantly less transparent and accountable: contributors or voting users face no consequences for getting it wrong and their methodology, bias, affiliation and funding is completely opaque. There is also no viable appeals or correction mechanism in place.
Fact checkers have repeatedly appealed to Meta to make the existing fact check labels more effective and less irritating by putting the information first instead of making users work to get to it. See the attached presentation which was shown to Meta in Sarajevo at Global Fact two years ago.