AI-generated 'news' site linked to OpenAI super PAC is smearing AI critics

An anonymously run outlet called Acutus (AcutusWire.com) presents itself as an expert-sourced wire service, but its reporters don’t exist. The site surfaced when the general counsel of AI-safety advocacy group Encode received an interview request from a ‘Michael Chen’ — a reporter with no online footprint, writing from a generic address, whose email an AI detector flagged as fully machine-written. Running all 94 of the site’s articles through the same detector, reporter Tyler Johnston found 69% fully AI-generated and 28% partially, with only three pieces judged human-authored. Johnston reports links between the operation and Targeted Victory, the firm behind OpenAI’s roughly $125 million political effort — the second such connection he’s traced this month.

The give-aways sit in plain view in the site’s own code. The React app ships its editorial back-end to every browser, exposing form fields like ‘AI Background Context’ and ‘Question Prompts’ plus ‘Generate Story Draft’ and ‘Regenerate’ buttons. A public API endpoint dumps the full story database along with each article’s automated review log: a multi-pass AI editor scores drafts on AP style, quote accuracy, and source verification, yet the median review runs just 44 seconds and publication follows about 10 seconds later. On 42 of the 94 stories, the system’s own reviewer marked the piece ‘needs_revision’ — and it was published anyway. A preserved ‘aiOriginalText’ field only makes sense if the drafts started as machine output.

The significance is less the automation than the intent. Quotes are largely scraped from existing web content, though a few real people were solicited via the fake-reporter pipeline, and articles have landed shortly before relevant legislation. Everything is licensed under Creative Commons and pushed out as a syndication feed, and the site is deliberately tuned for AI consumption — permissive robots.txt rules for AI crawlers, a ChatGPT plugin manifest, and an llms.txt boasting a ‘zero-hallucination editorial standard.’ The result is a laundering machine: synthetic, politically loaded content built to attack AI-industry critics and to be ingested and repeated by both humans and other models.