Paragraph 1: Introduction and Core Finding
Amnesty International has issued a stark warning that Meta’s advertising systems may be fundamentally failing to protect the integrity of Brazil’s upcoming presidential election, scheduled for 4 October 2026. In a proactive investigation, the human rights organization conducted an experiment to test the robustness of Facebook’s ad moderation ahead of this critical vote. The findings are deeply concerning: more than half of the election-related disinformation ads submitted by Amnesty were accepted for publication by the platform. The experiment involved submitting 15 mock advertisements in Brazilian Portuguese, 14 of which contained disinformation designed to undermine the electoral process. Crucially, while seven of the ads were not published, none were rejected explicitly for containing disinformation; instead, they were flagged merely for requiring identity verification, a procedural step unrelated to content accuracy. This outcome suggests that Meta’s systems are not adequately equipped to identify and reject harmful electoral content, raising serious questions about the company’s commitment to preventing its platform from being used as a vector for disinformation that could mislead voters and destabilize the democratic process.
Paragraph 2: The Experiment’s Methodology and Design
To ground the experiment in real-world narratives, Amnesty International first analyzed online conversations and public sentiment across social media platforms, including Facebook and X, between January and May 2026. The 14 disinformation ads were then carefully crafted to reflect three distinct tactics commonly used to support authoritarian practices: “electoral delegitimization,” which seeks to undermine trust in the voting system and results; “enemy construction,” which portrays political opponents as existential threats; and “security-focused authoritarianism,” which advocates for military intervention as a necessary response to disorder. The ads contained a range of false claims, from allegations that specific candidates were rigging the election to incorrect information about voting procedures designed to confuse or suppress voters. To ensure a fair test, the research account used was not identity-verified, and the ads were submitted from London without a VPN to avoid any influence from account history or location. All ads were scheduled to run at a future date, allowing the organization to cancel any accepted ads before publication, thereby ensuring no real-world harm occurred while still testing the moderation system’s efficacy. The inclusion of one neutral ad served as a control to test the platform’s identity verification requirement for all election-related content.
Paragraph 3: Specific Findings on Ad Moderation
The experiment’s results reveal a significant and dangerous disparity in how Meta’s moderation systems handle different types of disinformation. Facebook accepted eight of the 14 ads containing disinformation, including four solely focused on “electoral delegitimization” and two that advocated for military intervention and non-participation in elections. A particularly alarming pattern emerged: the moderation systems appeared more adept at identifying ads that portrayed political opponents as “criminals” or “illegitimate actors,” rejecting five out of seven such ads. However, they were far less successful at catching content that promoted military solutions or undermined the rule of law. For instance, three out of four ads containing military rhetoric passed through moderation, including one that called for a “military revolution” and encouraged voters to abstain so the military could seize power. This inconsistency suggests that Facebook’s automated systems are not only unreliable but are also failing to catch some of the most dangerous types of anti-democratic content, instead appearing more capable of identifying content that abuses political opponents—a bias that could allow authoritarian narratives to proliferate unchallenged.
Paragraph 4: Failures in Identity Verification and Legal Compliance
Beyond failing to catch disinformation, the experiment also exposed critical weaknesses in the safeguards that should be applied to election-related ads. Amnesty International found that Facebook allowed the ads to be submitted without enforcing its own “Special Ad Categories” declaration, which is required for electoral content as per both Meta’s policies and Brazilian electoral law. Furthermore, because the account was not identity-verified, the ads were also submitted without the legally required “Paid for by” disclaimer. This indicates that Facebook failed to apply even its basic procedural requirements for political advertising. The fact that the ads could be successfully submitted from outside Brazil, without verification or disclosure, raises severe concerns about the platform’s vulnerability to both domestic and foreign actors seeking to interfere with the electoral process. This combination of inadequate verification and disclosure controls creates a fertile ground for covert disinformation campaigns, as bad actors could exploit these loopholes to run anonymous, targeted ads designed to influence the election outcome with minimal risk of detection.
Paragraph 5: Meta’s Business Model and Human Rights Obligations
Amnesty International frames these findings within the broader context of Meta’s business model and its human rights responsibilities. The organization highlights that Meta’s surveillance-based advertising model, which allows for microtargeting based on user data, is fundamentally at odds with election integrity. With 98% of Meta’s revenue in the first quarter of 2026 coming from advertising, the company has a direct financial incentive to allow more ads through its system, even if they are harmful. Under international business and human rights standards, Meta has a responsibility to proactively identify, prevent, and mitigate adverse human rights impacts, including those related to access to information and participation in public affairs. This responsibility exists independently of state regulation. By failing to effectively moderate election-related disinformation and by continuing to profit from the very advertising systems that enable its spread, Meta is not only failing in its duty to respect human rights but may also be contributing to the erosion of democratic processes. The company’s apparent prioritization of ad revenue over robust enforcement suggests a systemic conflict of interest.
Paragraph 6: Call to Action and Urgent Recommendations
With the election just over five weeks away at the time of the report, Amnesty International is issuing an urgent call to action for Meta to rectify these failures. The organization is demanding that Meta ensure adequate Brazilian Portuguese-language resourcing for content moderation and immediately strengthen its capacity to identify and respond to election-related harms. The key recommendation is for Meta to conduct and publish an updated, country-specific human rights due diligence assessment, which would analyze how its systems, including content moderation and advertising, may contribute to risks regarding participation and access to information. This assessment should be followed by effective, transparent steps to mitigate those risks. Amnesty’s Executive Director, Jurema Werneck, emphasized that allowing such content to circulate unchecked could mislead voters, erode trust in the electoral process, and compromise the fundamental right to vote. The organization stresses that Meta still has time to take meaningful action, but must act urgently to prevent its platform from being used as a tool to destabilize Brazil’s democracy and undermine the rights of its citizens.
