Anthropic said on 14 October 2026 that it has strengthened Claude’s election safeguards ahead of the US midterms, adding official voting information and election-night results inside Claude.ai conversations and new classifiers aimed at multi-step election abuse.
In a sample of about 250,000 US Claude.ai conversations, Anthropic says about 5.75% touched on politics, public policy or social issues, while requests asking Claude whom or what to vote for showed up in about 0.01%.
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 organisation in this story.
The key insight: As we read it, the post is as much about where election answers come from as about what Claude refuses: Anthropic says official voting information from Democracy Works and up-to-date results will appear inside the conversation, so the chat window itself becomes a channel for official election information.
Anthropic says it used its Anthropic Insights tool to analyse a random sample of about 250,000 US Claude.ai consumer conversations from 21 to 27 September 2026, classified with what it calls a privacy-preserving method.
By Anthropic’s count, about 5.75% of US conversations on Claude.ai touch broadly on politics, public policy or social issues, and about 0.51% are specifically about elections or voting.
Requests that violated its election integrity policies were rarer still, at about 0.002% of conversations, Anthropic says. “As we get closer to Election Day, we expect the share of political and election-related queries to increase,” the company wrote.
Anthropic says about 0.01% of US Claude.ai conversations in a one-week sample of about 250,000 asked Claude whom or what to vote for, against about 5.75% that touched on politics, public policy or social issues.
Voting Information and Results Inside the Chat
Anthropic says Claude.ai will keep showing a banner that links to TurboVote, a nonpartisan resource from Democracy Works, when people ask about voter registration, polling places or election dates, as it first said in April.
Later this month, the company says, Claude will start drawing official voting information from Democracy Works into its answers for US users, so a question about where to vote or what ID to bring gets a state-specific answer without leaving the conversation.
Anthropic adds that Claude will begin showing up-to-date results on election night and in the days after, so that people asking for vote counts see the latest official numbers directly in their conversation.
The Bias Tests and What the Reviewers Found
Anthropic says Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 are its most evenhanded models to date, scoring 99% on its own political evenhandedness evaluation.
It asked three outside groups to review how Claude handles free expression: The Future of Free Speech, a think tank at Vanderbilt University; the Foundation for American Innovation; and the Collective Intelligence Project, which Anthropic says surveyed more than 2,000 people in the US, UK and India.
Anthropic reports that those reviews also found shortfalls. The Future of Free Speech and the Collective Intelligence Project both found that Claude rarely refused requests but “sometimes added unsolicited caveats and counterarguments, such as inserting an opposing view into an advocacy email.” The Foundation for American Innovation found that Claude Sonnet 4.5 “was more likely to soften arguments for right-leaning positions.”
The Rules for Campaigns
Anthropic’s Usage Policy prohibits using Claude to deceive voters or interfere with elections, the company says, from spreading false information about candidates or how to vote to impersonating election officials or trying to suppress turnout.
The company says it has published a guide for political campaigns. It lists permitted uses such as drafting press statements and fundraising emails, prepping for debates and analysing turnout data, and prohibited ones such as fake social media profiles posing as real people and content for news sites that falsely claim to be independent.
The Red-Team Tests and the New Classifiers
Anthropic says it worked with the Institute for Strategic Dialogue and the McCain Institute to pressure-test its protections in the US, Israel and Brazil, which all hold major national elections this year.
According to Anthropic, the analysts 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, the company says it added classifiers to catch novel and multi-step misuse, including attempts to build fake social media accounts and news sites aimed at political discussions, and strengthened detection of election-related misuse by foreign state actors.
What Anthropic Says It Has Already Caught
In one case, Anthropic says, foreign-based operators of a fake news site used Claude to produce US midterm coverage favouring Russian, Iranian and North Korean narratives. It also says it saw actors using Claude to produce fabricated news designed to attract American audiences and generate advertising revenue.
Anthropic says it banned the accounts involved in these cases. The same day it opened a Singapore office, and the US debate over AI in campaigns is also running at the FCC, which is weighing AI-voice political robocalls.
The Structural Read
As we read it, the usage figures cut against the idea that people hand their vote to a chatbot: by Anthropic’s count, about 0.01% of US Claude.ai conversations asked whom or what to vote for, against about 5.75% that touched politics broadly.
As we read it, the red-team finding is the operational point: Anthropic says outside analysts found Claude refused a single harmful request but could miss a campaign split across many harmless-looking requests, which is why the company says it added classifiers for multi-step misuse.
As we read it, publishing the reviewers’ criticism is part of the case Anthropic is making: the post reports both the 99% score on its own evenhandedness test and the outside findings that Claude added unsolicited caveats and softened right-leaning arguments.
Anthropic, on its outside red-team tests
“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”
Three Implications
VOTERS Anthropic says Claude.ai will draw official voting information from Democracy Works later this month and show up-to-date results on election night.
CAMPAIGNS Anthropic says it has published a guide setting out permitted uses such as drafting press statements and prohibited ones such as fake profiles posing as real people.
BAD ACTORS Anthropic says it banned the accounts behind a foreign-run fake news site and financially motivated fabricated news, and added classifiers for multi-step misuse.
The framework puts it this way: “It is helping define the rules, permissions, and boundaries around frontier intelligence.”
As we read it, the midterms post applies that to elections: Anthropic says it decides which political requests Claude answers, which it refuses under its Usage Policy, and which official sources it routes voters to.
Anthropic’s figures from a one-week sample of about 250,000 US Claude.ai conversations, 21-27 September 2026: about 5.75% touched politics, public policy or social issues; 0.51% elections or voting; 0.01% asked whom or what to vote for; 0.002% violated its election integrity policies.
What Is Not Established
Every usage figure in the post is Anthropic’s own measurement from one week of sampled Claude.ai consumer conversations; the post describes consumer conversations only, and it does not say that any outside party has audited the counts.
The 99% evenhandedness score comes from Anthropic’s own evaluation, and the post does not say how many accounts were banned in the cases it describes.
Business Engineer Framework
Inside Anthropic’s Permission Layer
A Business Engineer analysis of how Anthropic is turning safety rules, access and permissions into part of the AI stack itself.
Anthropic said on 14 October 2026 that it is adding official voting information and up-to-date election results to Claude.ai and new classifiers for multi-step election abuse before the US midterms.
By its own count, about 5.75% of US Claude.ai conversations touch on politics and about 0.01% ask Claude whom or what to vote for.
A note on sourcing. We read Anthropic’s post of 14 October 2026, “Preparing for the 2026 US midterms”, in full. Every usage figure, test result and enforcement case in this piece is Anthropic’s own account and measurement; we have not seen the underlying data or the reviewers’ reports. The Singapore office and the FCC item come from our own earlier reporting.
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, HackerNoon and Search Engine People. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
