Meta’s independent Oversight Board has repeatedly found the company’s approach to artificial intelligence-generated videos wanting. Shortfalls appear most stark in moments of political tension or armed conflict. The board’s March 2026 ruling on a fabricated clip about damage in Haifa, Israel, laid bare weaknesses that extend far beyond any single post. And the pattern keeps repeating.
Back in 2021 the board agreed to examine a case involving a deepfake video of a UK politician that Facebook had chosen to leave online. The Engadget report captured the moment when the quasi-judicial body first turned its attention to synthetic media impersonating elected officials. That early review signaled trouble ahead. Five years later the board still finds itself issuing the same core warning. Meta’s detection tools, labeling standards and enforcement thresholds fall short of what voters and viewers need.
The March 10, 2026, decision centered on an AI-created video posted from a Philippines-based Facebook account that presented itself as a news source. It depicted extensive destruction in the Israeli city of Haifa supposedly caused by Iranian strikes. None of it was real. The clip gathered nearly one million views. Users complained. Meta did nothing. No label. No removal. Only after the Oversight Board accepted an appeal did the company agree to apply a label within seven days.
“Meta must do more to address the proliferation of deceptive AI-generated content on its platforms… so that users can distinguish between what is real and fake,” the board stated in its ruling, as reported by the BBC. The advisers described the firm’s current methods as neither comprehensive nor fast enough to match the scale and speed of synthetic media, particularly during crises when platform engagement spikes.
That Israel-Iran video formed part of a larger wave. Pro-Israel and pro-Iran fabrications spread rapidly after hostilities escalated, some amassing more than 100 million views combined. The board determined that Meta’s bar for action — typically requiring risk of imminent physical harm — sits too high for content that can erode public confidence in all information. It recommended a new “high risk AI label” for such material and urged far more proactive labeling overall.
Months later the board selected another politically charged case. This one involved an eight-second clip of Hungarian opposition leader Péter Magyar, now the country’s incoming prime minister. Posted in November 2025 by an administrator of a politics-focused Facebook page, the video showed Magyar displaying exaggerated frustration over robocalls used in campaigning. Facial movements and speech patterns looked inauthentic. The post received more than 100,000 views and over 3,000 reactions, many of them laughing emojis.
Two hundred nine users reported the content, citing possible violations including misinformation and hateful conduct. Meta’s automated systems reviewed one report and decided the video did not breach the company’s misinformation policy. It did not qualify for an AI label, the firm said, because the clip appeared comedic, was posted well before the April 2026 elections, and did not seem tied to interference in political processes. The Oversight Board announced on April 23, 2026, that it would examine the matter to explore broader questions of AI in elections. Yet by May 12 the case evaporated. The original poster deleted the video, rendering board review impossible. Public comments gathered on the topic will still inform future decisions, the board noted in its official announcement.
These episodes reveal a consistent problem. Meta relies heavily on user reports, self-disclosure by creators, and after-the-fact fact-checking. During fast-moving events that system buckles. Overseas content farms churn out material designed to go viral. Political influencer networks, sometimes coordinated, amplify it. Satire blurs into deception. The board has asked for input on all these dynamics, from the incentives driving creators to the role of automated enforcement.
Earlier cases foreshadowed the difficulty. In 2024 the board criticized Meta’s manipulated media rules as incoherent after reviewing an altered video of then-President Joe Biden. The company left that clip up because it did not meet narrow criteria focused on AI alteration of video or audio. The board called for clearer standards. Meta has since expanded labeling requirements for political ads and rolled out tools such as Video Seal watermarking. Yet the Oversight Board’s recent opinions suggest execution still lags.
Recent reporting shows the stakes keep rising. A Guardian article from late March 2026 documented how military deepfakes and AI propaganda now carry real financial and political weight. The Oversight Board itself expressed concern that Meta applies labeling inconsistently, even to output from its own AI systems. Only a fraction receives proper disclosure.
Meanwhile scam advertisers exploit the same technology. Deepfake videos featuring politicians from both parties appear in Facebook ads promising nonexistent government rebates. These campaigns have drawn millions in ad spend, according to investigations that highlight weaknesses in pre-approval processes. The platform profits while users encounter increasingly convincing fabrications.
European institutions have sounded alarms too. The European Parliament highlighted Hungary’s use of unlabeled AI political content ahead of its elections. Similar worries surface in the UK, where deepfake smear campaigns targeted candidates during the 2024 general election cycle. Meta removed some networks after journalistic pressure, but reactive takedowns do not solve the underlying detection gap.
The Oversight Board, created by Meta in 2020, operates with unusual independence. Its 21 members issue non-binding yet high-profile recommendations. The company must respond publicly within 60 days. That mechanism has forced incremental policy shifts on everything from political speech to explicit deepfakes. On synthetic video the board now pushes for lower thresholds, better proactive tools, and clearer distinctions between satire and harmful deception.
Meta says it continues to refine its systems. It points to partnerships, open-source watermarking releases, and updated community standards. Yet the board’s language grows sharper with each decision. Current policies, it argues, cannot contend with the velocity of AI content in high-stakes environments.
So what comes next? The Hungarian case may be gone, but its questions remain. How should platforms weigh comedic intent against electoral risk? When does an exaggerated gesture in a short clip cross into misinformation? Can automated tools ever match human judgment at scale? And will Meta treat board recommendations on AI labeling with the seriousness required to restore some measure of trust?
Users already struggle to tell real from fake. That difficulty intensifies when videos of politicians or battlefield footage flood feeds during elections or wars. Without stronger action the risk is not only individual deception but a broader collapse in shared factual ground. The Oversight Board has issued its verdict. Implementation now rests with Meta. The record suggests change will arrive slowly, case by case, appeal by appeal.
WebProNews is an iEntry Publication