WASHINGTON, UNITED STATES — November 2, 2026. OpenAI and Anthropic have established dedicated election monitoring centers ahead of the November 3 U.S. midterm elections, marking the first time either artificial intelligence company has deployed a specialized operation of this kind. The initiatives are designed to identify attempts to misuse ChatGPT and Claude, respond to emerging misinformation threats and improve the reliability of election-related information. As millions of Americans prepare to vote, the companies are confronting a challenge that extends beyond technology: protecting democratic processes while ensuring that AI systems remain useful sources of information without becoming instruments of political manipulation.
According to reporting by Politico and Semafor, OpenAI established an Election Operations Center in late October, bringing together hundreds of employees from different departments. The operation, led by Josh Lawson, the company’s head of civic and strategic risk, is scheduled to remain active until the election votes have been counted. Its responsibilities include monitoring emerging incidents, investigating suspected violations of platform policies and responding to unexpected behavior by AI models. The objective is to coordinate internal expertise so that potential problems can be identified and addressed more quickly during a politically sensitive period.
Anthropic has introduced a comparable operation focused on Claude, its conversational AI platform. A company spokesperson confirmed to Semafor that dedicated personnel are monitoring potential misuse and preparing to respond rapidly to incidents involving election-related content. Unlike OpenAI, Anthropic has not disclosed the number of employees assigned to its monitoring center. Both initiatives reflect the growing recognition that conventional automated moderation systems may require additional human oversight when information spreads rapidly and inaccurate responses can have consequences for public understanding of electoral procedures.
The concept of an election war room is not entirely new. Meta introduced a similar system in 2018 to address concerns about disinformation and coordinated interference on social media platforms. However, generative AI presents additional challenges because chatbots produce individualized responses rather than simply displaying existing publications. These systems can generate text, images and other material that appears credible even when its underlying information is inaccurate. They can also be exploited to accelerate the production of fabricated news, impersonate legitimate sources or coordinate misleading communications across multiple digital channels.
Anthropic’s research provides insight into how Americans are using conversational AI during the electoral period. In an analysis of approximately 250,000 U.S. Claude conversations collected between September 21 and 27, the company found that 5.75% involved politics, public policy or social issues. Only 0.51% specifically concerned elections or voting, while requests asking the chatbot to recommend a political choice represented approximately 0.01%. Conversations violating the company’s election integrity policies accounted for about 0.002% of the sample. These findings suggest that most political interactions examined were informational rather than explicitly intended to influence voting behavior.
Nevertheless, the relatively low proportion of prohibited requests does not eliminate the possibility of coordinated abuse. Independent evaluations conducted with the Institute for Strategic Dialogue and the McCain Institute identified a significant vulnerability: Claude could reject an obviously harmful request while failing to recognize a broader deceptive operation divided into seemingly legitimate tasks. Such activities might involve researching audiences, creating websites, drafting misleading articles and preparing distribution strategies. Anthropic subsequently introduced additional detection mechanisms intended to recognize coordinated misuse across multiple interactions.
The company has also disclosed cases involving foreign-based operators who used Claude to produce fabricated political coverage. In one instance, individuals associated with a deceptive news website generated material promoting narratives favorable to Russia, Iran and North Korea. Anthropic reported disabling the accounts involved and strengthening its detection systems. However, the presence of content aligned with particular governments’ interests does not independently establish that those governments directed the operations. Other cases involved actors attempting to generate advertising revenue through fabricated election news, demonstrating that financial incentives can also drive disinformation.
Both companies are implementing additional safeguards to improve access to reliable electoral information. OpenAI announced on October 20 that ChatGPT had integrated information from Democracy Works, a nonpartisan organization that compiles voting requirements from state and local election authorities. Users can consult registration deadlines, voting locations and other practical information, while election-night vote counts are provided through The Associated Press after polling places close. Anthropic has announced similar arrangements involving Democracy Works and updated election results. Official electoral authorities, however, remain responsible for voting rules and the certification of final results.
OpenAI has also introduced measures addressing potentially misleading AI-generated imagery. Requests depicting election interference or voter suppression are directed toward stylized illustrations rather than photorealistic representations that could be mistaken for authentic documentary evidence. The company continues prohibiting certain forms of election interference and large-scale political persuasion campaigns conducted through its products. These restrictions are intended to reduce opportunities for deceptive manipulation while preserving legitimate uses such as research, administrative assistance and access to publicly available information.
The emergence of election monitoring centers also raises questions about transparency and institutional accountability. Technology companies must distinguish legitimate political discussion from coordinated deception, protect user privacy and avoid applying moderation rules inconsistently across ideological perspectives. At the same time, internal safeguards cannot guarantee that every inaccurate response or malicious operation will be detected. Independent research, public reporting and cooperation with election authorities remain important complements to company-operated monitoring systems.
As Americans approach the polls, the initiatives established by OpenAI and Anthropic illustrate how artificial intelligence is becoming increasingly embedded in the infrastructure of democratic communication. The immediate challenge is identifying misleading activity and correcting technical failures during a rapidly developing electoral process. The longer-term issue is determining how AI companies can provide accurate, politically neutral information while remaining accountable for the technologies they develop. Their performance during the 2026 midterms will provide important evidence for evaluating whether specialized monitoring operations can strengthen public trust in an increasingly automated information environment.
Phoenix24: Beyond the algorithms, the integrity of democracy. / Más allá de los algoritmos, la integridad de la democracia.