Report: Social media advances against disinformation, but AI chatbots lag

Social media and messaging applications, with the notable exception of Telegram, have made strides in combating electoral disinformation. However, artificial intelligence platforms are still failing to implement comprehensive policies to safeguard election integrity, according to a new report. The study, "The Role of Digital Platforms in Protecting Electoral Integrity," compiled by 65 research organizations, evaluated platforms based on their specific policies for the 2026 Brazilian elections and the transparency of their reporting channels.

“

AI platforms exhibit a very low or non-existent level of specific policies for electoral integrity.

YouTube and Meta's platforms, including Facebook and Instagram, received the highest ratings. They were followed by Kwai, WhatsApp, X, and TikTok. In contrast, AI platforms like Claude (Anthropic), Gemini (Google), and ChatGPT (OpenAI) are grouped together, showing similar deficiencies. MetaAI also trails behind. At the bottom of the rankings are Grok, X's chatbot, and the messaging app Telegram, neither of which has specific electoral policies in place.

“

AI platform policies are still in a very incipient stage. When they exist, they are difficult to find, are not clear, and in some cases are not even translated into Portuguese.

The report highlights significant concerns regarding AI chatbots' role in elections. "AI platforms exhibit a very low or non-existent level of specific policies for electoral integrity," the study states. While ChatGPT and Claude have policies for electoral integrity, they contain gaps. MetaAI, Grok, and Gemini lack specific rules for political AI use. Researchers note that AI platform policies are in a "very incipient stage," often difficult to find, unclear, and sometimes not even translated into Portuguese.

“

Anthropic (Claude) presents the most developed approach, combining a specific electoral integrity policy and a commitment already announced for informative banners for the 2026 Brazilian elections.

Anthropic's Claude is identified as having the most developed approach, combining a specific electoral integrity policy with a commitment to informative banners for the 2026 Brazilian elections. OpenAI offers the most comprehensive textual safeguards against electoral interference and fraudulent content, but its documentation focuses on user restrictions. Google, meanwhile, disperses information about Gemini across various institutional pages without a dedicated document on the tool's electoral governance. Meta and xAI are criticized for their opacity, with MetaAI and Grok lacking easily identifiable electoral policies and redirecting users to their parent companies' general policies.

“

OpenAI gathers the most comprehensive set of textual safeguards against electoral interference, voter suppression, and fraudulent synthetic content, but focuses documentation on user restrictions.