“The United States has no federal AI law. It has 50 state legislatures, all of them moving simultaneously, none of them coordinated, and a Congress watching from the sidelines while the deadline passes.”
Ground Truth · Episode 04 · June 2026On June 1st 2026, the New York State Legislature wrapped up its session by passing seven AI-related bills in a single day and sending them to Governor Kathy Hochul’s desk. The bills covered therapy chatbot disclosures, AI training data transparency, the FAIR News Act protecting publishers from AI scraping, a data centre construction moratorium, a ban on AI-assisted surveillance pricing, and children’s safety requirements for AI products directed at minors. Seven bills. One session. One state. In the same week, Rhode Island passed a full ban on AI therapy chatbots, and Colorado’s Governor Jared Polis vetoed a bill that would have prohibited algorithmic pricing. In Washington DC, a 269-page federal AI bill was introduced on June 5th. It has not moved out of committee. The result of all of this activity is, paradoxically, more confusion than clarity. Welcome to Episode 04 of Ground Truth, and welcome to the most important regulatory story in American technology that nobody is covering as a single story.
What Actually Passed, What Got Vetoed, and What Is Stuck
New York — Seven bills signed or pending Hochul’s signature. The package includes: a therapy chatbot disclosure law requiring AI mental health tools to identify themselves as AI; an AI training data transparency act requiring companies to disclose what data was used to train models deployed in New York; the FAIR News Act, which establishes a legal framework for publishers to seek compensation when their content is used to train AI without authorisation; a moratorium on new data centre construction pending environmental review; a ban on AI-assisted surveillance pricing in consumer markets; and a children’s safety bill covering AI products directed at users under 18. This is the most comprehensive single-session AI legislative package passed by any US state to date. It is also law only in New York.
Rhode Island — Therapy chatbot ban. The state became the first in the US to ban AI therapy chatbots outright rather than merely regulate them with disclosure requirements. The bill passed with bipartisan support following the widely reported cases of teenagers developing harmful emotional dependencies on AI companion apps. Rhode Island’s ban applies to any AI product marketed as providing mental health support, counselling, or therapeutic services. It is law only in Rhode Island.
Colorado — Governor Polis vetoed the algorithmic pricing bill. The bill would have prohibited the use of AI to coordinate pricing across competing businesses — a practice that critics argue amounts to algorithmic price-fixing without any human communication between competitors. Polis cited concerns about chilling legitimate dynamic pricing and said existing antitrust law was sufficient. Colorado’s AI Act anti-discrimination provisions, which require impact assessments for AI used in employment, housing, and credit decisions, take effect June 30th. That one was not vetoed. Companies with Colorado employees or customers have 17 days to comply.
Federal — The Great American Artificial Intelligence Act. Dropped June 5th at 269 pages, the bill proposes a national AI framework with a three-year preemption of conflicting state laws, a risk-tiered approach to regulation, and a new AI Safety Institute with enforcement authority. It also proposes preempting Colorado’s AI Act, New York’s package, and every other state law passed to date — retroactively. It has not been scheduled for a committee hearing. Several co-sponsors have already publicly distanced themselves from specific provisions. Congressional observers give it a low probability of passage before the November midterms.
June 13: Today. Colorado AI Act enforcement 17 days away. June 30: Colorado AI Act anti-discrimination provisions take effect. August 2: EU AI Act transparency obligations for chatbots and profiling systems. September 1: Several New York AI bills take effect if signed by Hochul. October 1: Additional EU AI Act provisions. No confirmed date: US federal AI Act. Currently: 0 federal AI laws specific to AI systems in the US. Current state laws active or pending: 50+ across 30+ states. Companies operating nationally: subject to all of them simultaneously.
This Is Not Regulatory Diversity. This Is Compliance Chaos.
The standard defence of state-level regulation is that it allows for experimentation and that well-designed state laws eventually become templates for federal action. California’s vehicle emissions standards, its consumer privacy law (CCPA), its data breach notification requirements — all became influential models for other states and eventual federal consideration. The argument is that states are laboratories of democracy and AI regulation benefits from that experimentation.
The argument has real merit at the level of policy design. It has serious problems at the level of compliance. A company deploying an AI hiring tool operates in all 50 states simultaneously. Colorado’s law requires algorithmic impact assessments. New York’s law requires training data disclosure. Illinois’ BIPA requires specific consent for biometric data used in AI systems and has generated over $650 million in settlements since 2019. California’s proposed AI liability framework would impose strict liability for AI-caused harms. Each of these laws defines key terms differently. Each has different notice requirements, different audit obligations, different enforcement mechanisms, and different penalty structures. Compliance with all of them simultaneously is not merely expensive. In several cases the laws are technically incompatible — what one requires, another prohibits or defines differently.
For large technology companies with substantial legal teams, this is painful and expensive. For the mid-market businesses — the HR software company, the healthcare analytics firm, the financial services provider — that have integrated AI into their core products without the resources to run parallel compliance programmes across 50 jurisdictions, it is an existential operational challenge. And for the startups building the next generation of AI applications, the compliance overhead of the patchwork is beginning to function as a barrier to entry that protects incumbents without necessarily improving safety.
The companies best positioned to navigate a 50-state patchwork of AI regulation are the largest companies with the most sophisticated legal teams. The companies most harmed by it are the startups and mid-market firms the regulation was partly designed to protect people from. Regulatory complexity that protects incumbents is not consumer protection. It is incumbent protection wearing consumer protection’s clothes.Neal Lloyd · Ground Truth, Episode 04
Europe Has One Law. It Has Problems. But At Least It Has One Law.
The EU AI Act is not a perfect piece of legislation. It was drafted over three years during which the technology moved faster than the drafters could keep up with. Its risk-tiered approach — prohibited, high-risk, limited-risk, minimal-risk — produces classification questions that lawyers have been arguing about since the text was finalised. Its extraterritorial scope means it applies to any company deploying AI that affects EU citizens, regardless of where the company is based. Its compliance burden on small operators is real and contested. It is being criticised from the left for not going far enough on transparency and from the right for not going far enough on innovation facilitation.
But it is one law. It applies uniformly across 27 countries and 450 million consumers. A company that achieves EU AI Act compliance has a single compliance posture for the entire European market. The definitions are consistent. The enforcement body is identifiable. The obligations are, at least in principle, knowable in advance. Compare this to the US situation — where a company must track active legislation across 50 states, comply with whichever provisions have taken effect, manage the uncertainty of which federal preemption bill if any will pass, and do all of this while the technology continues to evolve faster than any of the laws can track — and the EU’s approach, with all its flaws, starts to look considerably more functional.
The August 2nd EU deadline for chatbot transparency and profiling system obligations is 50 days away. OpenAI, Anthropic, Google, and Apple are all actively managing EU compliance programmes. The US companies deploying AI have a compliance calendar for Europe. They do not have an equivalent clarity for their home market. That is a striking statement about the relative functionality of the two regulatory environments.
A 269-Page Bill, Zero Hearings, and an Election in Five Months
The Great American Artificial Intelligence Act is the most ambitious federal AI legislation ever introduced in the United States. It is also, by the assessment of most Washington observers, unlikely to pass in its current form before the November 2026 midterm elections. The bill’s three-year preemption of state laws has already alienated the state AGs and consumer advocacy groups whose support would be needed to build a coalition. Its enforcement mechanism — a new AI Safety Institute inside the Commerce Department — faces jurisdictional objections from the FTC, the FCC, and financial regulators who have all claimed AI oversight authority in their respective domains. The 269 pages contain enough contested provisions to sustain opposition from both sides of every relevant debate.
Congress’s AI track record is not encouraging. The landmark AI legislation most observers consider most likely to pass in 2026 is a narrow, bipartisan bill requiring disclosure labels on AI-generated content in political advertising — a genuine but extremely limited intervention in a field that requires considerably broader governance. The gap between what the situation demands and what the legislature is positioned to deliver has been a consistent feature of American technology policy for thirty years. There is no structural reason to expect it to close faster for AI than it did for social media, for data privacy, or for algorithmic accountability.
If your company uses AI in hiring, lending, or healthcare and you have customers or employees in Colorado: you have 17 days to complete an algorithmic impact assessment or face enforcement action. If you deploy AI products to children in New York: review your compliance posture against the children’s safety bill immediately. If you use AI to generate content for New York news consumers: the FAIR News Act is coming. If you are an EU operator of any AI chatbot that builds user profiles: August 2nd is your hard deadline. If you are waiting for federal preemption before taking state compliance seriously: the preemption bill is not coming on your timeline. Act on the state laws as written.
The US approach to AI regulation is not laissez-faire. It is 50 different approaches, none coordinated, several contradictory, all simultaneously in force, with a federal solution that is 269 pages long and has not had its first committee hearing. That is not less regulation. That is worse regulation.Neal Lloyd · Ground Truth, Episode 04
Ground Truth, Episode 04 · June 2026
Neal Lloyd covers the real-world impact of AI — money, power, geopolitics, and the stories behind the headlines. Ground Truth is his daily AI news and analysis series on emdexter.blogspot.com.
