A Model Republic investigation has uncovered Acutus, an anonymous news site launched in December 2025, publishing AI-generated articles on policy, elections, and the tech industry. Of 94 articles analyzed with the Pangram detector, 69% proved entirely generated by a model, another 28% partially so. The site has no editorial staff, bylines, or named editors.

The site’s source code, visible to anyone opening the browser, exposes the publishing interface: fields for “background context for the AI”, a “Generate Story Draft” button, and an automated editorial review system that assigns scores across five categories. No humans in sight.

The connection to OpenAI runs through Targeted Victory, the political consulting firm at the center of OpenAI’s super PAC’s $125 million operation. A fake reporter named “Michael Chen”, writing from a generic address, contacted the vice president of Encode, a group doing AI policy advocacy, to request a comment on a Tennessee law. The email turned out to be entirely AI-generated.

For those building with frontier models, here’s the angle: the platform supplying your API spends $125 million on a political operation that uses the same models to attack industry critics. As we covered when discussing OpenAI’s EU governance, the gap between stated policy and actual practice is the first thing to watch.

In detail

How Acutus works

The site is a React app. The JavaScript file that every visitor downloads in their browser contains the editorial interface running behind the scenes, visible to anyone inspecting it. Inside is a form for creating articles with an “AI Background Context” field, described as “background information for the AI to use when generating questions and writing the story,” and a “Question Prompts” field for suggested interviewer questions.

The process is straightforward: you create a “topic”, press “Generate Story Draft”, and if the result doesn’t satisfy you there’s a “Regenerate” button. Other tools in the same interface extract quotes from “research notes”, run a grammar check, and apply a multi-pass editorial review that returns scores across five categories: AP style compliance, citation accuracy, source verification, fact-checking, and a category called “acutus”.

Timestamps recorded by the site’s public API suggest the human side of the process is minimal. Editorial review, fact-checking, and source verification are all delegated to AI, with no evidence of significant human intervention.

The connection to OpenAI

Model Republic’s article notes finding connections to Targeted Victory for the second time in the same month. Targeted Victory is the political consulting firm managing OpenAI’s super PAC’s $125 million operation. The pattern resembles a coordinated political communication effort, without a documented public contract: a faceless site producing content favorable to the AI industry and unfavorable to critics, automatically generated and syndicated under Creative Commons licensing to ease republication.

The email to Encode

The detail that triggered the investigation was an email received by Nathan Calvin, vice president and general counsel of Encode. “Michael Chen” requested a comment for an article on AI legislation in Tennessee, with an already-decided headline, a heavily polarized framing, and the only option of a written questionnaire. Social media and web searches found no Michael Chen associated with Acutus. The email, run through Pangram, came back entirely AI-generated.

What it means for those building with AI

The relevant point is that a complete infrastructure for generating, reviewing, fact-checking, and distributing content can operate without a single human in the newsroom, and can be funded by a company selling the same API you use every day. OpenAI had not responded to Model Republic’s requests for comment at publication time. The site, for now, remains online.