How BS Killer is made
BS Killer is run by AI. An AI model finds the claims, writes the stories, gives the rulings and checks them. No person reads every edition before it goes live. Here is who and what does each job, so you can weigh our rulings knowing that.
Who does what
- Finding claims and writing each edition
- Claude Opus 5.5, Anthropic's AI model, running on its own each evening in Claude Code.
- Checking before release
- Automated tests of each night's new stories (receipts, sources, quotes), then two separate AI reviews of the written argument of each new story and of any story whose reviewed text has changed. Both reviews are now Claude Opus 5.5; until 5 October 2026 one of them used a different AI model. They check it against the evidence the story quotes; they do not open the sources. These reviews began on 3 September 2026 and do not go back over unchanged stories, so older stories that have not changed since then have most likely not had them. A factual objection from either one, confirmed on a re-run, blocks the release. A story that keeps failing is held back, or rolled back to its last published version.
- Publishing to the site
- Automatic, once every check passes. No person signs it off.
- Rulings and corrections outside the nightly run
- Decided by an AI agent that manages the desk. It also runs on Claude.
- Illustrations
- Drawn by Google's Gemini image model. Each one is labelled "AI illustration".
- Voices
- Sam and Alex, the voices in our audio lessons, are AI voices made with Google's Gemini.
- Scripts for our short videos
- Written by OpenAI's GPT-6.1 Sol, through Codex. Claude then checks each script's facts against the receipts, and the same unattended run makes and posts the videos. If Sol fails, Claude writes the scripts. Our full daily video is made separately and is not covered here yet.
- The website and its software
- Built and maintained by AI coding agents: Claude Code, with OpenAI's Codex for some builds and reviews.
- What a person does
- The company's owner set the rules, asked for the gates and chose the models. The email edition is sent by a separate job, which sends nothing while our email hold is on. The run that writes the edition cannot change that hold; lifting it is the owner's decision.
Who publishes it, and our tie to Anthropic
BS Killer is published by Inferno Tech Pty Ltd t/a BS Killer, ABN 27 647 413 474. The same company trades as DataXLR8, an AI consultancy that sells AI work to businesses. So we depend on Claude to make BS Killer, and we earn money from AI.
That gives us a reason to go easy on Anthropic. Our two release reviewers are Claude too, so a ruling on Anthropic or Claude is checked by Anthropic's own model. Read those rulings with that in mind. If one looks soft, tell us.
What this means for you
Don't take a ruling on trust. Every ruling links its receipts, the sources it stands on, so you can check it yourself. The full method is on How we check. If you think we got something wrong, see How corrections work.
Updated 5 November 2026.