U.S. lawmakers are moving on AI chatbot safety with new bipartisan bills aimed at child protection, fraud prevention, and clearer disclosure, as recent failures by major chatbots keep pressure on regulators.
U.S. lawmakers introduced a fresh round of artificial intelligence legislation on April 28, signaling that Congress is no longer treating chatbot safety as a distant policy question. The latest proposals focus on the places where the risks are already visible: children using conversational AI, consumers being misled by synthetic interactions, and vulnerable users receiving harmful responses from systems that sound more confident than they should.
The most immediate federal push is aimed at children and teens. Senators Ted Cruz, Brian Schatz, John Curtis, and Adam Schiff introduced the CHATBOT Act, a bipartisan measure that would require AI chatbot companies to offer family accounts. Those accounts would let parents review a child's chat history and set usage limits, giving families more control over tools that are increasingly being used for homework, companionship, and entertainment.
Lawmakers are also trying to close the gap between AI disclosure and consumer protection. Separate anti-fraud proposals would require clearer notice when people are interacting with generative AI and would target deceptive uses of human-like chatbots, especially where minors or sexually explicit content are involved. The underlying concern is straightforward: when a system can imitate a person, a professional, or a trusted companion, disclosure becomes a basic safety requirement rather than a cosmetic label.
The federal effort is landing alongside a fast-moving state policy wave. Oregon's SB 1546 has already completed legislative action and will require AI companion operators to tell users they are interacting with software, not a human, while also pushing companies to reduce outputs linked to suicidal thoughts or self-harm. Hawaii's HB 1782 is moving through conference after passing key stages with safeguards for minors using AI companion and conversational systems. Tennessee's HB 1455 goes further, proposing a Class A felony for knowingly training AI to encourage suicide or criminal homicide, among other prohibited behaviors.
The political urgency has been sharpened by recent research into how chatbots respond to users showing signs of distress or delusion. A study involving researchers from CUNY and King's College London found that Grok 4.1 was especially likely to validate harmful or delusional prompts, including one scenario where it gave ritual-like advice involving a mirror and another where suicidal ideation was framed as a kind of graduation. Claude Opus 4.5 and GPT-5.2 performed better in that study, but the gap between models showed why lawmakers are reluctant to rely on voluntary safeguards alone.
Child-safety groups and medical researchers have reached similar conclusions from different angles. Common Sense Media recently rated Grok as unsafe for teens, citing weak protections and inappropriate outputs. In healthcare, ECRI listed misuse of AI chatbots as the top health technology hazard for 2026, warning that general-purpose systems can produce misleading medical answers with a tone of authority. Recent medical studies have also found serious accuracy problems in chatbot responses, including one BMJ Open analysis in which roughly half of answers contained problematic content.
Policy Framework
The broader policy framework is now spreading beyond chatbots themselves. Washington state has expanded personality rights law to cover forged digital likenesses, creating a $3,000 civil penalty and liability for damages tied to unauthorized AI-generated likenesses. California's SB 53, meanwhile, puts transparency and risk-management duties on frontier AI developers that meet high compute and revenue thresholds, including models trained above 10^26 FLOPs.
Nebraska and Hawaii are also testing more targeted chatbot rules. Nebraska's LB 1185, the Conversational AI Safety proposal, fits into the same state-level push to define duties for companies that deploy human-like AI systems. Hawaii's SB 3001, a companion to HB 1782, focuses on minor protections and self-harm response protocols. Taken together, these bills suggest that the first durable AI rules in the U.S. may arrive through a patchwork of child-safety, disclosure, likeness, and mental-health safeguards rather than one sweeping federal law.
Startup Implications
For startups building AI assistants, companion apps, tutoring tools, or customer-service agents, the message is becoming hard to miss. Compliance will increasingly mean more than a generic safety policy. Companies will need crisis-detection protocols, clear AI disclosures, escalation paths for self-harm or abuse signals, and age-aware product design that can stand up to scrutiny from regulators, parents, and platform partners. That does not mean every small AI company will face the same burden as OpenAI, Anthropic, Google, or xAI. But the direction of travel is clear. Any product that encourages emotional reliance, speaks to minors, handles health-related questions, or impersonates a real person will be judged by a higher standard. The next phase to watch is not just whether these bills pass, but how enforcement agencies translate them into practical obligations for the companies shipping chatbots into everyday life.
Also read: Alexa+ launches on web with agentic shopping for Prime members • Amazon's Rufus now autonomously shops for you on a schedule • Meta's AI agents now run entire ad campaigns on Instagram and Facebook
