As Americans vote in the 2026 midterm elections today, 3 November, OpenAI and Anthropic have each set up "war rooms" to monitor their AI systems for misuse and wrong information, according to reporting by Politico and Semafor. It is the first time either company has done so for an election, Semafor reported. Here is what the companies say they are doing, and what experts say the limits are.
OpenAI's Election Operations Center
OpenAI told Politico that its Election Operations Center, announced on 28 October, brings together staffing "in the hundreds" and will run until the votes are counted. It is led by Josh Lawson, the company's head of civic and strategic risk. Its job is to monitor abuse of ChatGPT in real time, handle reports of unexpected model behaviour or suspected policy violations, and point users to partners for accurate information. OpenAI's election safeguards page, updated on 20 October, lists several measures:
- Voting information from the nonprofit Democracy Works, and live vote counts from the Associated Press in ChatGPT after polls close, for US and Brazilian elections.
- A feature the company calls cartoonification, which turns photorealistic election-related image requests into cartoon-style output.
- Provenance markers on ChatGPT-generated images, and no political advertising this cycle.
- Special access to cyber-defence models for eligible election authorities and voting-system manufacturers.
Anthropic's approach
Anthropic published a post titled "Preparing for the 2026 US midterms" and told Semafor that it has a dedicated war room monitoring its platform for misuse. The company says it has banned users for election-related abuse, including operators of a fake news site pushing Russian, Iranian and North Korean viewpoints, and people using Claude to produce fabricated news for advertising revenue. Anthropic says fewer than 1% of Claude conversations are about elections or voting, a share it expects to rise as election day approaches. It also says Claude will show up-to-date official results on election night and the following days when users ask for vote counts.
What experts caution
Katie Harbath, a former Meta public-policy director, told Semafor that war rooms are built to handle things that cannot be predicted and that need a quick, established process for triage. Semafor noted that AI's unpredictability, recent cybersecurity incidents and the tendency of AI search tools to reflect results users want to see all make the job harder. Some analysts have pointed out that safeguards on image tools have in the past been bypassed with simple follow-up prompts, and say that how quickly companies can escalate and act matters as much as how many people they assign. Senator Mark Warner and several Democratic colleagues wrote to major technology companies last month urging them to protect the integrity of the midterms. Microsoft has also said malicious deepfakes of candidates and election officials are disallowed on its platforms.
For voters
AI chatbots can make mistakes on time-sensitive facts. For anything practical, such as where and when to vote, what ID you need, or whether a result is official, rely on your state or local election office and official government websites rather than a chatbot, and be sceptical of unverified images, videos and screenshots shared on social media today.
Sources: Semafor (2 November 2026); Politico, via AI Weekly and FourWeekMBA; OpenAI election safeguards page; Anthropic's midterms announcement, via NewsNation; Nextgov/FCW; eWeek.
