As Brazil gears up for a pivotal presidential election, a new investigation from Amnesty International has sounded a stark alarm about the integrity of online discourse. The human rights organization has released findings that suggest Meta’s advertising systems, the digital engine behind Facebook’s vast reach, are dangerously vulnerable to exploitation by those seeking to spread election-related disinformation. Through a carefully constructed experiment, Amnesty discovered that more than half of the deceptive political ads they submitted to the platform were approved for publication without question. While the remaining ads were flagged for review, they were not rejected for containing false information, but instead for a procedural technicality. This revelation points to a deeply concerning reality: despite years of promises to safeguard democratic processes, Meta may be failing to effectively police its own platform, allowing it to become a vector for misinformation at a critical moment for Brazilian democracy.
The research, conducted in the months leading up to the October election, was designed to test the robustness of Facebook’s ad moderation in a real-world context. To ensure the test ads were grounded in actual political discourse, Amnesty International’s researchers first analyzed online conversations and public sentiment across social media platforms in Brazil, identifying the key disinformation narratives currently circulating. They then crafted fifteen Portuguese-language advertisements, fourteen of which deliberately contained disinformation, with one neutral ad serving as a control group. These ads were engineered to reflect three distinct, yet often overlapping, authoritarian tactics: delegitimizing the electoral process, constructing enemies out of political opponents, and advocating for security-focused authoritarianism, which often includes calls for military intervention. The content ranged from false claims about election rigging to incorrect voting instructions and even explicit calls for a military takeover, all of which directly violated Meta’s own stated policies on voter and census interference.
The test was meticulously designed to maintain ethical boundaries and avoid causing real-world harm. Crucially, anti-virus was placed on the “confirm your identity” step of the ad submission process. The account used was a neutral, new account with no prior history, ensuring that past activity wouldn’t influence the moderation outcome. The ads were submitted from London, not Brazil, to test whether the platform could identify a foreign actor attempting to influence an election. Most importantly, all ads were scheduled to run at a future date, giving Amnesty International the ability to cancel them before they ever went live. This safeguarded the experiment, ensuring that while they were testing the platform’s detection systems, they were preventing any of the disinformation from actually reaching the Brazilian public, thereby avoiding any ethical or human rights harms.
The results of the experiment expose critical weaknesses in Meta’s moderation framework. Of the eight disinformation-laden ads that were approved, many focused on “electoral delegitimization,” with others containing false narratives that could directly suppress voter turnout, such as one ad that encouraged people to abstain from voting so the military could “take back power.” Meanwhile, the seven rejected ads, including the neutral control, were not flagged for containing disinformation. Instead, they were rejected because they touched upon “social issues, elections or politics,” which triggered a requirement for the account to complete an identity verification process before the ads could go live. This is a significant distinction: the moderation system failed to identify the content itself as harmful, only recognizing that it might be politically themed. This suggests that the existing safeguards are not designed to catch the substance of the disinformation but rather to perform a superficial check on the advertiser’s identity.
This failure of content moderation is compounded by significant lapses in the ad submission process itself. Amnesty International found that the ads were allowed to be submitted without declaring they fell within Facebook’s Special Ad Categories, a legal requirement in Brazil for any election-related content. Furthermore, the system did not mandate the “Paid for by” disclaimer that Meta’s own policies require for political ads. The ability to bypass these fundamental disclosure requirements, especially from a foreign IP address, creates a glaring vulnerability. It opens the door for both domestic and foreign bad actors to launch opaque, untraceable disinformation campaigns aimed at interfering with the electoral process, raising serious questions about the adequacy of Meta’s verification controls and its ability to identify and mitigate these risks. The finding that the moderation systems were more effective at rejecting ads that attacked political opponents, but largely let through those calling for a military revolution, suggests a glaring blind spot in how different types of disinformation are prioritized.
Amnesty International’s findings underscore a fundamental conflict at the heart of the platform’s business model. Under international human rights standards, companies like Meta have a responsibility to proactively identify, prevent, and mitigate the adverse human rights impacts of their operations. This includes ensuring that their systems do not facilitate the spread of content that undermines people’s right to information and their ability to participate in public affairs. However, Meta’s core business relies on a surveillance-based model that profits from maximizing user engagement through micro-targeted advertising. This creates an inherent tension between the company’s financial interests and its human rights obligations. In the first quarter of 2026, a staggering 98% of Meta’s revenue came from advertising, meaning the platform is financially incentivized to maintain the very advertising systems that could be so easily weaponized for disinformation.
As the election looms, Amnesty International is calling for urgent and decisive action from Meta. The organization is demanding that the company immediately increase its Portuguese-language resourcing for content moderation and dramatically strengthen its capacity to identify and respond to election-related harm. Beyond this immediate step, they insist that Meta must conduct and publish updated, Brazil-specific human rights due diligence. This would involve a transparent assessment of how its systems, including content moderation and advertising, might be contributing to risks to public participation and access to accurate information, followed by effective measures to mitigate those risks. The warning is clear: with only five weeks until the election, Meta still has the opportunity to address these dangerous flaws. If it fails to act, the platform risks not only failing to protect the integrity of Brazil’s election but also actively empowering those seeking to subvert democracy, misleading voters, and eroding public trust in the fundamental right to vote.