The US House of Representatives has imposed strict limits on the use of ChatGPT in its offices. In fact, it is a limit aimed above all at the staff, not only of the institution, but also of the parliamentarians themselves. The approved rules reflect similar limits decided by numerous private companies, in the USA and in the rest of the world, and are based on the fear that conversations with the chatbot may not be safe: not only because the company, at least if it is not set otherwise, uses them to train and improve its technology, but also because nothing prevents companies such as OpenAI or Google from being exposed to cyber attacks, with the risk that the information provided to ChatGPT ends up in the hands of malicious people.

According to a memo obtained by Axios, staff can only use the paid ChatGPT Plus service for “research and evaluation.” Furthermore, the staff of parliamentarians can only provide the chatbot with data accessible to the public: it follows that it is forbidden to provide ChatGPT with inputs that include reserved or confidential information.

Currently, the free version of ChatGPT and other large language models are not allowed to be used. This policy aims to avoid problems such as the creation of laws or AI-generated speech. Two lawyers and a law firm were recently sentenced to pay a $5,000 fine after they used ChatGPT to generate a legal document, which turned out to be littered with fabricated information and citations of non-existent laws and precedents. Both sides of Congress are working to regulate AI, with bills requiring caveats for the use of generative AI and increased scrutiny for political ads.