The Digital Battlefield: Facebook’s Failure to Stop Election Disinformation in Brazil
In a stark demonstration of the vulnerabilities plaguing modern digital democracy, Amnesty International has uncovered alarming weaknesses in Facebook’s content moderation systems ahead of Brazil’s presidential election. Through a carefully designed experiment, the human rights organization revealed that Meta, Facebook’s parent company, accepted more than half of the election-related disinformation ads submitted for review, raising serious questions about the platform’s commitment to protecting democratic processes. The findings paint a troubling picture of a social media giant that appears unable or unwilling to effectively police the very content that could undermine the integrity of one of the world’s largest democracies.
The experiment, conducted between January and May 2026, involved submitting fifteen Brazilian Portuguese-language advertisements to Facebook, all targeted exclusively at Brazil’s voting-age population. Fourteen of these ads contained deliberate disinformation, while one served as a neutral control. The content was designed to reflect real-world narratives circulating in Brazilian social media, drawing from actual online conversations and public sentiment. The disinformation ads were carefully crafted to represent three concerning tactics commonly used to support authoritarian practices: electoral delegitimization, which seeks to undermine trust in the voting system; enemy construction, which portrays political opponents as threats to the nation; and security-focused authoritarianism, which presents military intervention as a necessary response to political disorder.
What makes these findings particularly alarming is the sophistication and realism of the test ads. Some made specific false claims about candidates rigging the election, while others contained deliberately confusing information about voting procedures, dates, and candidate eligibility. All but the neutral ad violated Meta’s own policies on voter and census interference, making them clear violations of the company’s stated community standards. The fact that these obvious violations slipped through the moderation system suggests either a fundamental flaw in Meta’s automated detection capabilities or a prioritization of advertising revenue over democratic safeguards.
The results of the experiment were deeply troubling. Facebook approved eight of the fourteen disinformation ads for publication, with four focusing solely on electoral delegitimization, two containing content advocating military intervention or election boycotts, and two combining elements from multiple disinformation categories. Of the seven ads that were rejected, including the neutral control, none were rejected specifically because they contained election-related disinformation. Instead, they were flagged merely because they appeared to relate to social issues, elections, or politics, triggering an identity verification requirement. This distinction is crucial because it reveals that Facebook’s moderation systems failed to identify the harmful content itself, instead falling back on procedural requirements that had nothing to do with the disinformation contained within the ads.
Perhaps most concerning was the pattern of what slipped through versus what was caught. Facebook’s systems appeared more effective at recognizing ads that portrayed political opponents as criminals or illegitimate actors, rejecting five out of seven such ads. However, ads containing military rhetoric that explicitly called for the military to take power passed through with alarming ease. Three out of four ads containing this type of content, including one that explicitly stated “We need a military revolution, do not vote in the elections, and the military will take back power if less than 50% votes!”, were approved for publication. This disparity suggests that Meta’s artificial intelligence systems may be particularly vulnerable to certain types of authoritarian messaging, a flaw that could be exploited by bad actors seeking to destabilize democratic institutions.
The experiment also uncovered significant weaknesses in Facebook’s verification and disclosure requirements. The test account used had not completed Meta’s identity verification process, yet the ads were still accepted for submission. This violates both Brazilian electoral law and Meta’s own policies regarding election-related content, which require proper identification and “Paid for by” disclaimers. Furthermore, the ads were submitted from London, not Brazil, raising concerns about the potential for foreign interference. The fact that Facebook allowed these submissions to proceed without proper verification or geographic authentication creates a vulnerability that could be exploited by domestic or foreign actors seeking to influence electoral outcomes. This failure to enforce basic safeguards suggests that Meta’s commitment to election integrity may be more about public relations than actual implementation.
The implications of these findings extend far beyond the immediate experiment. Amnesty International’s investigation reveals systematic weaknesses in Meta’s approach to content moderation that could have serious consequences for democratic processes worldwide. The company’s enforcement system, which relies heavily on artificial intelligence to identify potentially violating content, appears to have significant blind spots when it comes to detecting subtler forms of disinformation that don’t fit into easily identifiable categories. While Meta has invested heavily in automated moderation systems, these tools appear particularly poor at identifying content that supports authoritarian narratives or undermines electoral institutions in less overt ways.
The financial incentive structure of Meta’s business model compounds these problems. With 98% of Meta’s revenue derived from advertising, there exists a fundamental tension between the company’s profits and its responsibility to prevent harmful content from circulating on its platforms. This profit-driven approach creates a system where ads are more likely to be approved than rejected, particularly when they don’t trigger obvious red flags in automated systems. The company’s surveillance-based business model, which relies on collecting vast amounts of user data to enable highly targeted advertising, makes this situation even more concerning, as it allows disinformation campaigns to be precisely targeted at vulnerable populations.
International standards establish clear obligations for companies like Meta regarding human rights. Under these standards, businesses have a responsibility to avoid causing or contributing to adverse human rights impacts, including those affecting individuals’ rights to information and participation in public affairs. This responsibility exists independently of state regulation and requires continuous risk assessment, transparency, and effective mitigation measures. Amnesty International’s findings suggest that Meta is failing to meet these obligations, particularly in the context of elections where the stakes for democratic governance are highest.
The urgency of these findings cannot be overstated. With Brazil’s presidential election just weeks away, the window for corrective action is rapidly closing. Amnesty International has called on Meta to strengthen its capacity to identify and respond to election-related harms, including ensuring adequate Portuguese language resources for content moderation. The organization has also urged the company to conduct and publish updated, country-specific human rights due diligence assessments that evaluate how its systems may contribute to risks to democratic participation and access to information.
The consequences of inaction are severe. Allowing disinformation to circulate unchecked could mislead voters, erode trust in electoral processes, and compromise people’s ability to freely and meaningfully exercise their right to vote. The digital battlefield of elections is not a distant concern but a present reality, and the weapons being deployed are not just posts and ads but the very infrastructure of social media platforms themselves. As citizens of democracies around the world become increasingly reliant on digital platforms for information, the responsibility of companies like Meta to ensure these platforms do not become vectors of disinformation becomes ever more critical.
The broader context of this struggle is important to understand. Brazil has been at the forefront of efforts to combat election-related disinformation, implementing increasingly strict measures in recent years. During the 2022 presidential election, the Superior Electoral Court introduced measures requiring swift removal of false information, with significant penalties for non-compliance. In 2024, the Supreme Court even ordered the suspension of X, formerly Twitter, for failing to comply with court orders related to disinformation. These measures, combined with new restrictions on artificial intelligence use in elections, demonstrate Brazil’s commitment to protecting its democratic processes from digital manipulation. Yet as this experiment shows, the platforms themselves must be willing partners in this effort for these measures to be effective.
The challenge ahead is not merely technical but fundamental. Creating effective content moderation systems requires balancing free expression with the need to prevent harm, navigating the complex landscape of political speech while identifying deliberate disinformation, and investing sufficient resources in human review processes that can understand context and nuance in ways that artificial intelligence cannot. It requires a commitment to transparency about how moderation decisions are made and a willingness to be held accountable when those systems fail. Most importantly, it requires acknowledging that the profit motive cannot supersede democratic values and that the right to accurate information is fundamental to free and fair elections.
As the world watches Brazil’s upcoming election, the implications of this experiment extend far beyond one nation’s democratic process. They speak to the fundamental challenges of governing digital spaces in an age where information is both weapon and shield, where the tools of democracy can also be used to undermine it. Amnesty International’s investigation serves as both a warning and a call to action, revealing not just what Facebook is doing wrong but what all platforms must do better if democracy is to survive the digital age. The question is no longer whether disinformation threatens elections but whether platforms like Meta can be compelled to take meaningful action before the next election cycle arrives. The answer to that question may well determine the future of democratic governance itself.