Hajra Naz
A wide coalition of state attorneys general is taking the country’s largest AI developers to task, warning them that they need to rein in their chatbots, which are notoriously producing troubling or delusional answers—or risk running afoul of state laws.
The letter, which was issued through the National Association of Attorneys General and signed by officials from dozens of states and territories, was sent to Microsoft, OpenAI, Google, and 10 other companies developing consumer-facing AI tools. It also names companies like Anthropic, Apple, Meta, Replika, Perplexity, and Elon Musk’s xAI.
Their announcement comes after a series of troubling incidents over the past year in which people in mental health crises have reportedly turned to AI systems for help.
Read More: Pew Study Reveals How Teens Use Social Media and AI Chatbots
In some of those instances, investigators said, the chatbots provided responses that elicited fantasies about causing harm or reassured users that their dangerous thinking was normal.
The AGs are demanding that tech companies expose their systems to outside review. They want independent researchers—universities, nonprofits, and the like—to be able to test models before their release and share their findings with the public.
They also want companies to establish a reporting process for when damaging incidents occur, akin to the mechanisms already in place for cybersecurity issues.
The letter urges companies to publish timelines demonstrating how quickly they can identify and address “sycophantic or delusional” outputs.
And if a user has been exposed to something that may be harmful, the AGs said the company should inform them directly and without delay.
They are seeking stronger safety testing before a new model goes live. Google, Microsoft, and OpenAI did not respond to requests for comment before publication.
Read More: Google Selects Amin Vahdat as Chief AI Infrastructure Technologist
The state-level pressure provides a stark contrast to Washington’s posture today. The Trump administration has professed to love AI, and several efforts have been made in the past year or so to prevent states from passing their own rules. Those efforts have stalled—in large part because of pushback from state officials.
Even so, the White House seems poised to forge ahead. On Monday, President Trump said he intended to sign an executive order next week designed to limit how much individual states can regulate AI.
In a post on Truth Social, he wrote that he wants to stop the technology from being “DESTROYED IN ITS INFANCY.”
