Anthropic has set up an election “war room” for the first time, assembling teams of analysts in cells in San Francisco, New York and Washington, The New York Times reported on 3 November, the day of the US midterm elections. The Times said OpenAI is running its own operations center for the vote, and that Google and Meta have also pledged efforts to stop their AI tools being used to undermine confidence in the democratic process.
Anthropic’s own figures from 14 October give a sense of the scale of what such teams look for. In a one-week sample of 250,000 US conversations on Claude.ai, it said requests that violated its election integrity policies appeared in about 0.002% of conversations. That is about 2 in every 100,000, or about 1 in 50,000, by our arithmetic, while conversations about elections or voting were about 255 times as common, at about 0.51%.
According to the Times, OpenAI stood up its election center last Thursday in a conference room at one of its San Francisco offices, where a team of more than a dozen people works around the clock, with a satellite center in Washington; the effort involves scores of people, company officials told the paper. The team examines ChatGPT prompts that have been automatically flagged for attempts to create false texts or images about the election, and the company notifies other platforms when it finds malign activity intended for their sites, the Times reported.
Josh Lawson, OpenAI’s head of civic and strategic risk, who leads the center, told the Times: “The most available technology at the time in 2024 was text”. He added: “We’re not just monitoring text anymore”. OpenAI’s chief investigator, Ben Nimmo, said: “We’ve been trying to catch operations … before they have a chance to gain traction”. The Times reported that OpenAI has not yet disclosed any specific efforts in the midterm elections.
The Times also cited the Brennan Center for Justice, which found that the most popular chatbots generally pushed back on queries about election-related conspiracy theories, while AI tools regularly created text, images and videos that could be used to undermine confidence in elections.
Conversations per 100,000 in Anthropic’s one-week sample of 250,000 US Claude.ai conversations, by topic, converted from the percentages Anthropic stated on 14 October 2026 (our arithmetic). Requests that violated Anthropic’s election integrity policies were about 0.002%, about 2 in 100,000. Categories overlap and all shares are approximate, as Anthropic states them.
Business Pill · Guardrails
A one-minute explainer of guardrails: rules enforced around an AI model rather than inside it. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read the sources, election-day defence at the AI companies now has a physical form: the Times reports Anthropic’s first war room, with cells in San Francisco, New York and Washington, while Anthropic’s own October figures put rule-breaking election requests at about 0.002% of US Claude.ai conversations, about 1 in 50,000 by our arithmetic.
What Anthropic Says Its Teams Look For
Anthropic’s 14 October post says it deploys automated classifiers trained to detect election-related abuse. It says testers from the Institute for Strategic Dialogue and the McCain Institute found that “Claude reliably refused a single harmful request, but could miss a malicious campaign when it was spread across many individual requests that looked harmless on their own”.
In response, Anthropic says it added classifiers aimed at novel and multi-step misuse, including attempts to build fake social media accounts and news sites aimed at political discussions. It describes one case in which “foreign-based operators of a fake news site used Claude to produce US midterm coverage favoring Russian, Iranian, and North Korean narratives”, and says it banned the accounts involved in that case and in a case of financially motivated fabricated news.
The same post says about 5.75% of US Claude.ai conversations in the sample touched broadly on politics, public policy or social issues, and that requests asking Claude whom or what to vote for showed up in about 0.01%. Anthropic wrote: “As we get closer to Election Day, we expect the share of political and election-related queries to increase”.
How the Four Companies Describe Their Election Measures
OpenAI’s election safeguards page, updated on 20 October, says its Democracy Works integration is live in the US. It says its integration with The Associated Press brings live vote counts into ChatGPT after polls close, and that it applies “cartoonified” stylization to requests for images depicting election interference or that may discourage voting. It says: “We continue to prohibit political ads from running on our platform”.
Google’s 9 September post says it is rolling out election tools in Search, AI Mode, AI Overviews and the Gemini app, drawing on state and local governments and Democracy Works, with real-time results from The Associated Press. It says its Trust & Safety team and Google Threat Intelligence Group work around the clock on cyberattacks and coordinated influence operations aimed at elections.
Meta’s election post, updated on 27 October, says it has “activated our Election Operations Center in the final week before Election Day in the US” and began a Restriction Period on new political, electoral and social issue ads that day. The post says Meta has removed 200 networks of Coordinated Inauthentic Behavior since 2017 and has spent more than $30 billion on safety and security over the last decade.
By our count of the four companies’ own posts, three of them, OpenAI, Anthropic and Google, name Democracy Works as a source of voting information inside their products. OpenAI and Google name The Associated Press for live results; Anthropic’s post says Claude will show “the latest official numbers” without naming a provider. Meta’s post points people to state and local election officials.
Why It Matters for the Companies
The measures sit alongside policy positions. OpenAI’s page says it supports two Senate bills, the Protect Elections from Deceptive AI Act (S. 1213) and the Preparing Election Administrators for AI Act (S. 2346). Anthropic’s amended IPO filing of 2 November carries a risk factor warning that its political engagement could bring heightened scrutiny or political backlash.
As we read it, the small share and the new war room point the same way. Anthropic’s testers found that Claude could miss a campaign spread across many requests that looked harmless on their own, so the rare violations are the visible part; the multi-step operations are what the classifiers and analysts are set up to find.
As we read the four posts, the companies have converged on the same voting-information plumbing: by our count, three of the four name Democracy Works inside their products, and OpenAI and Google both name The Associated Press for live results. Where they differ is in what they publish about misuse: Anthropic gives a share of conversations, Meta a count of networks removed since 2017.
Josh Lawson, OpenAI, to The New York Times, 3 Nov 2026
“We’re not just monitoring text anymore”
Three Implications
FOR TRUST Three of the four companies, OpenAI, Anthropic and Google, tie voting-logistics answers to Democracy Works, and OpenAI and Google name The Associated Press for results; Meta points people to state and local officials, by their own posts.
FOR REGULATIONOpenAI’s page backs two Senate bills on deceptive AI and election administration, and Anthropic’s IPO filing lists political backlash as a risk factor, by their own documents.
FOR DISCLOSURE Only Anthropic, among the four posts we read, gives a share of conversations that broke its election rules; the Times reported that OpenAI has not yet disclosed specific midterm efforts.
The analysis reads: “The governance layer sits at the intersection of capability, regulation, national security, access control, and geopolitical alignment”.
As we read it, an election war room is that governance layer at work on one day: decisions about who gets to use a model, and for what, made by analysts and classifiers rather than by regulators. The analysis was published in June 2026 and does not discuss the midterms.
We read the Times article to its paywall. The description of Anthropic’s war room, its cells and its first-time status is the Times’s; in the documents we read, Anthropic’s 14 October post does not describe a war room, and we did not find an Anthropic statement on it. The size of Anthropic’s teams is not established.
Anthropic’s shares come from one week of US Claude.ai conversations and are stated as approximate, so the per-100,000 rates and the 255-to-1 ratio, which are our arithmetic, are approximate too; the categories overlap. Neither OpenAI, Google nor Meta has published a comparable share in the posts we read. This piece makes no forecast about the election; results come from AP race calls and official returns.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Anthropic has set up an election war room for the first time, with cells in San Francisco, New York and Washington, the Times reported, joining OpenAI’s operations center and Meta’s Election Operations Center on election day. By Anthropic’s own October figures, requests that broke its election rules were about 0.002% of US Claude.ai conversations, about 1 in 50,000 by our arithmetic.
A note on sourcing. The war room details are from The New York Times, which we read to its paywall. Company measures and figures come from each company’s own election post: Anthropic (14 October 2026), OpenAI (updated 20 October), Google (9 September) and Meta (updated 27 October). Per-100,000 rates, the 1-in-50,000 and 255-to-1 figures and the count of companies naming Democracy Works are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
