A digital news site that looks like a legitimate publication is sending AI-generated story pitches to real journalists and policy advocates, and the operators don’t appear to be human. According to a report surfaced on Hacker News, Acutus (acutuswire.com) launched on December 29, 2025, and has published 94 full-length articles on AI policy, Senate races, nuclear energy, crypto regulation, and more, all without a masthead, bylines, or named editors. The Hacker News story also notes that OpenAI appears to be funding the operation, which raises uncomfortable questions about who’s bankrolling synthetic journalism.
The story broke when Nathan Calvin of advocacy group Encode received an email from ‘Michael Chen,’ a reporter requesting comment on a Tennessee AI bill. Multiple red flags appeared immediately: a generic reporter@acutuswire.com address, heavily loaded framing, ‘written Q&A only’ as the sole format, and zero web presence for the supposed reporter. Pangram, an AI content detector with a near-zero false-positive rate, flagged the email as fully AI-generated.
What’s actually running Acutus
Investigators ran every article on the site through Pangram with damning results:
- 69% flagged as fully AI-generated
- 28% flagged as partially AI-generated
- Only 3 articles classified as human-authored
The evidence isn’t just statistical. Acutus’s React app exposes its editorial interface in the public JavaScript bundle. Anyone who inspects the source code can see fields labeled ‘AI Background Context’ and ‘Question Prompts’ feeding an AI interviewer, plus a ‘Generate Story Draft’ button that produces full articles. There’s also a ‘Regenerate’ option and a multi-pass AI editorial review that scores pieces against editorial benchmarks.
The site’s /api/wire endpoint goes further, returning the entire production database including AI editorial review results. Categories include AP style compliance, quote accuracy, source verification, and fact-checking, all scored out of 100 by the AI itself.
The numbers tell a brutal story
Timestamps in the API expose how little human review actually happens. The full multi-pass editorial review completes in a median of 44 seconds. Publication happens a median of 10 seconds after the last flagged issue is resolved.
Even more damning: on 42 of the 94 stories, the AI’s own overall quality check returned ‘needs_revision.’ Every single one was published anyway.
Why this matters
Acutus licenses everything under Creative Commons as a ‘wire service,’ meaning other publishers can pick up its content without payment. That’s the danger. Synthetic articles with synthetic sources can launder into real outlets through syndication, then get cited as primary reporting elsewhere. This is exactly how a polluted information ecosystem propagates.
This isn’t an experiment. It’s a working pipeline targeting policy debates, regulatory questions, and political races. The investigator traced field names like ‘aiOriginalText’ in the code, which only make sense if AI is doing the original writing while humans (if any) rubber-stamp the output.
What practitioners should do
If you’re a journalist, advocate, or comms professional, treat unsolicited interview requests as suspect by default:
- Run inbound emails through an AI detector before responding
- Verify reporters have a real digital footprint outside the publication
- Be skeptical of ‘written Q&A only’ pitches with pre-loaded framing
- Check whether the publication has bylines, an editorial team page, and clear ownership
- Refuse to comment until you can confirm a human is on the other end
The bigger picture for the AI industry is uglier. If a major lab is underwriting this kind of operation, it signals that model providers are willing to fund synthetic media that pollutes the very information ecosystem they helped create. Expect more sites like Acutus to surface before platforms or regulators catch up, and expect them to get harder to spot.
Full technical breakdown and screenshots are available at the original Hacker News thread.