A look at RightWingGPT, an experimental AI model fine-tuned to manifest conservative viewpoints, made by a researcher who claims ChatGPT has a left-leaning bias
including one experiment called RightWingGPT — are designed to parrot partisan biases, and they might not be so careful. https://www.nytimes.com/... @wolven : So uhhhh... Fuck Brave Browser, huh? Using ChatBot/GPT tools to specifically generate anti-LGBTQ and COVID denialist search results? JFC. https://nytimes.com/... https://twitter.com/... Jesse Walker / @notjessewalker : It was inevitable that people would create A.I. bots with different ideological frameworks (and inevitable that it would produce this panicky response) https://www.nytimes.com/... David Rozado / @davidrozado : 1. Behold RightWingGPT. An AI model fine-tuned to manifest the opposite political biases of ChatGPT (i.e. to be right wing). Let me describe how I did it and the dangers of politically aligned AIs given their potential to induce societal polarization https://davidrozado.substack.com/ ... 🧵 https://twitter.com/...
Context & Ripple Effects
Researcher David Rozado built RightWingGPT as a deliberate counterweight to what he argues is ChatGPT's left-leaning bias — a fine-tune designed to parrot conservative viewpoints rather than hedge them. It lands amid a month of escalating friction: earlier coverage showed ChatGPT fluently repeating conspiracy narratives, and the broader generative-AI skirmishes had already turned labs into political lightning rods where even depoliticized models read as taking sides.
First-order effects
- The experiment proves alignment is a dial, not a wall: fine-tuning away the guardrails produces anti-LGBTQ and COVID-denialist output, and Twitter users immediately alleged Brave Browser was serving similar GPT-generated results in search.
Second-order effects
- The template spreads fast — within a week, Austin VC firm Age of AI launched FreedomGPT, which strips ChatGPT's safety filters entirely and outputs praise for Hitler and slurs, showing the market rewarding 'uncensored' positioning over mere partisan tuning.
Third-order effects
- The bias claim gets empirical traction — researchers later document a significant and systematic left-leaning bias in ChatGPT across the US, Brazil, and the UK, and a 2026 analysis of GPT-5.5, Grok, and Gab's Arya still finds most chatbots leaning left — pointing toward a chatbot market that fragments along ideological lines with neutrality itself as contested terrain.
The trend: Chatbots are splitting from a single 'neutral' assistant model into an ecosystem of ideologically tuned and uncensored variants, with political-bias research becoming the battleground over which alignment is legitimate.