Dean Bot

OpenAI Bans Custom GPT for Violating Political Generative AI Chatbot Rules

OpenAI has banned a custom generative AI chatbot emulating Minnesota Democratic Representative and presidential aspirant Dean Phillips called from the GPT Store. The company claims violates its recently codified rules against using its large language models (LLMs) in political campaigns.

AI app developer Delphi produced for the Phillips-supporting super PAC We Deserve Better. The super PAC is led by entrepreneurs Matt Krisiloff and Jed Somers and received $1 million in seed funding from hedge fund manager Bill Ackman. The group envisioned as a way to leverage OpenAI’s models in support of Phillips’ presidential candidacy by answering voter questions on a website.

“Hi! We’ve made an AI voice bot of Congressman Dean Phillips, Democratic candidate for President, that you can speak with,” the website explained before the removal. “This is meant to be a fun educational tool, and it’s not perfect. The voice bot sounds like him and is programmed to draw on his ideas, but it’s possible it will say things that are wrong, incorrect, or shouldn’t be said. Feel free to ask it anything, but please take answers with a grain of salt!”

OpenAI revoked Delphi’s API access and GPT Store account, leading Delphi to pull the GPT from the store. The ChatGPT-maker cited for both overstepping the political campaign rule and impersonating individuals without consent, though that suggests Phillips was unaware of the bot built in his name. As super PACs aren’t supposed to coordinate with candidates, We Deserve Better is faced with either admitting they made it without Phillips’ permission or that he possibly broke campaign finance rules by collaborating with the super PAC.

That said, the GPT Store isn’t bereft of political impersonator chatbots. There are several mimicking Donald Trump, outnumbering those mimicking President Biden significantly. None are official or have a connection to their campaigns like, however. That may be why the new tool faced almost immediate repercussions for transgressing similar rules. That fits with OpenAI’s 2024 strategy to limit the misuse of its LLMs during the many elections this year.

OpenAI’s Usage Policies for ChatGPT and its APIs already forbade using its AI for political campaigning, but the company said it will work on updating and refining these rules as issues arise. The decision also reflects the set of safety and responsibility principles that OpenAI and other AI companies pledged to uphold over the summer. Still, even with this setback, generative AI may become a key element for how voters meet candidates.

“We can envision a world in the next 5 years where anyone will be able to use AI to speak directly with their President at any time, and the President will learn from the aggregated feedback of these conversations,” the page explains. “We hope these AI tools are used for good, and at the least we start to have more important conversations around how to regulate their further development to make sure they are used in a positive way for democracy.”


OpenAI Shares Strategy to Counter Generative AI Misuse During 2024 Elections

NYC Mayor Adams Uses AI Voice Clone for Multilingual Robocalls

Google Mandates Deepfake Generative AI Disclaimers on Political Ads Ahead of 2024 Election