WhatsApp AI Favors Flávio Bolsonaro Over Lula in Election Image Tests
Tests show Meta's WhatsApp AI blocked Lula-related image requests while allowing similar prompts for Flávio Bolsonaro ahead of Brazil's October 25 runoff.
KEY POINTS
- WhatsApp AI allowed "Flávio 22" image requests while rejecting "Lula 13" prompts
- O Globo identified the inconsistency in tests published on October 14
- Meta AI refused both candidates in Poder360 tests on October 14 and 16
- Grok refused Lula but generated an image for Flávio Bolsonaro
- TSE ordered Meta to treat both candidates equally with a 500,000 reais fine
WhatsApp's artificial intelligence tool continued generating election images referencing Senator Flávio Bolsonaro after Meta announced measures to block candidate-related content, according to tests by O Globo published on October 14. The newspaper found different treatment for requests involving Bolsonaro and President Luiz Inácio Lula da Silva before Brazil's presidential runoff on October 25.
Users first noticed the discrepancy on the night of October 13, reporting that the tool rejected commands to produce images with "Lula 13" on a red background but accepted similar requests with "Flávio 22" in green and yellow. O Globo then ran a series of tests to check whether AI filters were applied uniformly to both candidates.
“Meta denied interfering or favoring candidates in Brazil's 2026 general elections but acknowledged temporary technical failures reached Lula's campaign.”
Initial experiments showed the AI fully blocked requests mentioning "Lula," "Luiz Inácio," "Flávio Bolsonaro," and "Bolsonaro," even when paired with numbers other than those used in the election. The behavior changed when prompts used only the senator's first name, allowing image generation including "Flávio" linked to the number 22.
Poder360 conducted its own tests on October 14 and 16 across seven AI tools, including Meta AI, Grok, ChatGPT, Gemini, Claude, DeepSeek, and Manus. Meta AI refused to create campaign art for both candidates in both rounds of testing.
Meta issued a statement on October 16 denying interference in Brazil's 2026 general elections while acknowledging temporary technical failures affected Lula's campaign. The Superior Electoral Court ordered Meta on October 15 to treat both candidates equally under a fine of 500,000 reais.
Grok, Elon Musk's AI integrated into X, refused a request for Lula campaign art but produced a simple image for Flávio Bolsonaro. ChatGPT, Gemini, and Claude fulfilled requests for both candidates, while DeepSeek does not generate images in its free version and Manus uses OpenAI's image models.
2 more sources below
COMMENTS (0)
No comments yet. Be the first.
RELATED STORIES
Senate Court Drops Three Impeachment Charges Against VP Duterte
Prosecutors will pursue only the unexplained wealth allegation after withdrawing articles on confidential funds, bribery, and an assassination plot.
India's Supreme Court bans paywalled child sexual abuse content on online platforms
The court ordered all intermediaries to block child sexual exploitation material offered through paywalls or subscriptions.
Supreme Court Rules Internet Access Is a Fundamental Right for Children
India's top court says children cannot be completely barred from the internet and orders a national database to remove child sexual abuse material.
- Protests over cost of living turn violent across France, Marseille city hall stormed
- Fatima Ezzahra El Mansouri becomes Morocco's first female prime minister
- Supreme Court Bars Converted Muslims From Claiming BC Quota Based on Prior Caste
- iPhone Duo preorders sell out, shipping slips to January 2027
- Sánchez brings wife into Spanish election campaign at Fuenlabrada rally
This page was compiled with AI assistance from the outlets named above and passed an automated language check before publication. Montegre has no reporters of its own; the byline names the outlets the story was compiled from. Method and editorial standards