OpenAI and Anthropic have each set up dedicated ‘‘war rooms’’ focused on ensuring their platforms provide accurate information during the upcoming U.S. midterm elections. This is the first time either company has deployed such a unit.
The concept of a tech platform ‘‘war room’’ dates back to 2018, when Meta established one amid rising concerns about the spread of election-related misinformation. The arrival of advanced AI has amplified those threats: new risks include deepfake videos and AI-driven search results that can propagate incorrect information.
A recent example underlining the problem: last week, Google’s AI Mode declined to answer a query about who won the 2020 election; Google said it was working to fix the issue. Incidents like this show how errors in AI systems can directly affect public information.
An Anthropic spokesperson told Semafor: “Teams across the company work year-round on strengthening our election-related safeguards and enforcing our policies. Ahead of the election, we have a dedicated war room monitoring our platform for misuse and ready to quickly respond to any incidents.”
Running such war rooms is complicated by AI’s unpredictable behavior, recent cybersecurity concerns, and the tendency for AI search tools to mirror what users expect or want to see in results. Katie Harbath, former director of public policy at Meta, told Semafor that these war rooms are ‘‘built for dealing with the things we can’t necessarily predict going wrong and you need a quick, established process in terms of triaging.’’
These measures indicate growing attention from tech companies to the integrity of election-related information, while also highlighting the ongoing challenges of managing new, AI-driven risks.
