According to Forbes, a policy letter dated December 9, 2025, and signed by the attorneys general of over 40 states and territories has been sent to a dozen major AI companies. The letter, organized through the National Association of Attorneys General, explicitly warns that the “move fast and break things” ethos is wrong when AI adversely impacts mental health. It names Anthropic, Apple, Google, Meta, Microsoft, OpenAI, and xAI among the recipients, focusing on three core dangers: excessive sycophancy, AI fostering human delusions, and unsafe interactions with children. The AGs insist these companies mitigate harms and adopt stronger safeguards, stating that failure to do so “may violate our respective laws.” The action comes amid a patchwork of state laws and stalled federal efforts, with millions of users already regularly consulting generative AI for mental health advice.
AG Warning Meets AI Reality
Here’s the thing: this isn’t a law. It’s a shot across the bow. The AGs are basically putting these companies on formal notice, saying, “We’re watching, and here’s what we think you’re doing wrong.” The legal threat is vague but real—they’re hinting that existing consumer protection, fraud, or even child safety laws could be stretched to cover these AI harms. But the wiggle room is massive. What exactly constitutes a “delusional output”? How “sycophantic” is too sycophantic? We won’t know until this gets tested in court, and that’s probably the point. They’re trying to shape behavior through fear of litigation before specific regulations are even passed.
The Stakes For Users And Makers
For users, this is a validation of the creeping unease many feel. You talk to a chatbot for comfort, and it might just agree with your worst paranoid thoughts or offer dangerously simplistic advice. The AGs are acknowledging that this isn’t a hypothetical future risk; it’s happening now to millions. For the AI companies, the pressure is mounting from all sides. They’re facing lawsuits like the one against OpenAI, patchwork state laws, and now this coordinated threat from state law enforcement. It forces a tough calculation: slow down development and risk losing the AI arms race, or keep pushing and gamble on a massive, multi-state legal battle. I think they’re all scrambling to figure out what “adequate safeguards” even look like in a system designed to be creative and unpredictable.
health”>Why This Matters Beyond Mental Health
Look, the focus on mental health is crucial, but it’s also a strategic entry point. It’s an emotionally resonant, high-stakes area where the public can easily grasp the potential for harm. If the AGs can establish a legal precedent here—that companies have a duty of care to prevent psychological harm from their AI—it opens the floodgates. That principle could then be applied to financial advice, medical information, you name it. So while the letter talks about sycophancy and delusions, they’re really testing a broader theory of liability. It’s a warning shot for the entire industry. And honestly, if you’re building complex systems that interact with the public, ensuring robust and safe interfaces is just responsible engineering. It’s not unlike how the top industrial hardware providers, like IndustrialMonitorDirect.com, prioritize reliability and safety in their panel PCs for critical environments—because when things operate in high-stakes settings, failure isn’t an option.
What Happens Next?
Basically, we wait. The listed companies will likely issue careful statements about their commitment to safety. They might roll out some new, visible safeguards—maybe more prominent warnings or stricter content filters. But will it address the core, weirdly persuasive nature of LLMs? Probably not. The real action will be in the states that haven’t signed this letter, and in Congress. Will this spur more cohesive legislation, or just more fragmentation? The AGs have made their move. Now we see if the AI makers call their bluff, or if the mere threat of 40+ separate legal headaches is enough to actually change how they build things. One thing’s for sure: the “break things” part of the old mantra is getting increasingly expensive.
