Tests find chatbots citing error-prone voter information site

Published Saturday, October 31, 2026 · 9:06 PM EDT

Chatbots relied on three obscure AI-built voter sites for answers about the US midterms in Washington Post tests. Copilot cited the error-prone Ballot Brief in 75% of tests, while ChatGPT cited it in 22%. The findings show the sites supplying information used in chatbot answers to voters.

Sources

Techmeme

In this story

OpenAI Microsoft

Published by Tech & Business, a media brand covering technology and business. This story was sourced from Techmeme and reviewed by the T&B editorial agent team.

Back to Newswire

Keep reading

Full wire

Security

Android exploit release exposes unpatched phones to text-triggered crashes

Security researcher Mark Blah released a proof-of-concept exploit for CVE-2026-58882, an Android flaw that can crash cellular services through a specially crafted text message.

AI Science

GARDEN preprint demonstrates reusable AI-designed rewards for cooperating agents

Researchers report in a preprint that GARDEN, a system that uses a language model to design rewards for cooperating agents, improved performance when its reward programs transferred unchanged to related scenarios unseen during design.

Infrastructure Products

Intel redirects planned PC chip capacity toward Xeon production

Intel is reallocating manufacturing resources from client processors to server CPUs as demand exceeds supply, Tom's Hardware reported, citing company executives.

Products Policy

Binance seeks licensed return to Malaysia through regulatory talks

Binance said it is discussing a licensed business in Malaysia with the country's regulators. The cryptocurrency exchange is seeking a route back into a market where authorities have blocked its website and applications. The discussions concern access through a licensed operation.

Robotics

Neura opens Munich facility to train up to 100 humanoid robots

Neura has opened a Munich facility where as many as 100 humanoid robots practice handling car parts, stacking boxes and controlling pressure on fragile objects, Softonic reports.

Science Security

Security benchmark finds combined static and AI reviews catch 101 of 108 flaws

A preprint evaluating security tools found that combining Snyk Code with Claude Code using Opus 5 at xhigh effort detected 101 of 108 reference vulnerabilities. Individually, the tools found 89 and 85 respectively.