Kathmandu— US congressional lawyers are facing an increasing influx of flawed, AI-generated legislative proposals, requiring significant revisions before they can be considered viable, according to a report by Politico. Staffers and outside groups are increasingly utilizing tools like ChatGPT and Claude to draft legislation, but these drafts often contain incorrect statutory references, confused legal definitions, and potentially unintended consequences if enacted. The House Office of Legislative Counsel (OLC) is struggling to keep up with the volume of requests, which has risen sharply in recent months, while also addressing the errors within the AI-generated text. This issue extends beyond Capitol Hill, with similar mistakes appearing in Canadian legislatures and US court rulings.

Rising Demand, Mounting Errors

The OLC received 5,623 legislative requests during the first 60 days of the current Congress – a 72% increase compared to the same period two years prior. This surge in demand is compounded by the need to meticulously review and correct AI-drafted legislation. One individual advising congressional staff on the technology noted that lawyers are often spending more time fixing these drafts than it would take to create them independently.

The errors range from incorrect statutory references and confusing legal definitions to wording that could lead to unintended consequences. In one instance, an AI draft defined “state” in a way that potentially excluded Washington D.C. or tribal nations from federal programs. The technology also struggles with nuanced legal mechanisms like tax credits, deductions, and grants.

Public Blunders and Expanding Use

The use of AI in legislative work is becoming increasingly common on Capitol Hill, with four congressional staffers confirming to Politico that chatbots are used for both reviewing and drafting legislation. However, this increased reliance has already resulted in several public errors. Representative Anna Paulina Luna’s office published an amendment summary in June containing the phrase “Claude responded:” alongside text copied directly from the chatbot.

A similar incident occurred in Canada when New Brunswick lawmaker Bill Oliver read leftover instructions from a chatbot aloud during a legislative speech, including the prompt: “Here's a more natural, flowing version of that section that reads like a legislative speech rather than a series of short points.”

Broader Implications Beyond Congress

The issues with AI-generated content extend beyond the US legislature. Two federal judges faced scrutiny last year for rulings containing inaccurate quotations and factual errors, with AI reportedly involved in at least one case. The State Department also issued an apology after using an AI-generated map that incorrectly placed several African countries.

Despite these challenges, the OLC is exploring potential benefits of AI, forming a working group to examine its use for legal research and introducing technology to assist lawyers in drafting legislation.

The OLC continues to assess how artificial intelligence can be integrated into its workflow while simultaneously addressing the risks associated with inaccurate or poorly drafted legislative text. The increasing prevalence of AI tools suggests that managing these challenges will remain a key concern for lawmakers and legal professionals.

(With inputs from RT)

Originally published on abcnews.com.np.