Michigan AG vs. US AI Law
The regulation of artificial intelligence (AI) in the US is the scene of a power struggle between states and the federal government. 36 attorneys general, led by Michigan Attorney General Dana Nessel, are warning the US Congress against taking away the states' ability to enact their own AI laws. This conflict highlights the fundamental question of whether a uniform federal law or a patchwork of state regulations will determine the future of AI governance in the US.
Source: Own Representation
Introduction
The Attorney General of Michigan and 35 other attorneys general from US states and territories have sent a joint letter to the leadership of the US Congress. In it, they are calling for a halt to a proposed bill that would impose a nationwide moratorium on state-level AI regulation. At its core, the issue is whether uniform, rather industry-friendly rules from Washington should apply, or whether states should be allowed to enact their own, sometimes significantly stricter, AI laws, for example, concerning consumer protection, discrimination, or deepfakes.
The Conflict in the US
Dana Nessel, the Michigan Attorney General, speaks in her statement of an attempt to effectively exclude the states from AI regulation for years. Together with 35 other attorneys general, she is calling on Congress not to pass any clause that would prevent states from enacting or enforcing their own AI laws. The letter, published via the National Association of Attorneys General (NAAG), , is addressed to the four most important party leaders in Congress: Speaker Mike Johnson, Majority Leader John Thune, and Minority Leaders Hakeem Jeffries and Chuck Schumer. The signatories warn of a "moratorium" on state AI laws, which in their view would primarily benefit large tech corporations – but not consumers, children, or employees who are already confronted with AI scams, deepfakes, or opaque decision-making systems.
Source: lgbtqnation.com
Dana Nessel, the Attorney General of Michigan, is a prominent voice in the debate surrounding the regulation of artificial intelligence.
What is described in the letter as an "AI moratorium" is not an entirely new project. As early as May 2025, an amendment appeared in the US House of Representatives that would have imposed a ten-year moratorium on all state laws concerning artificial intelligence and automated decision-making systems. Michigan was already involved at that time: Nessel joined a coalition of 39 attorneys general who criticized exactly this ban as "far-reaching" and "destructive" to sensible state AI regulation. The amendment was attached to a large budget and tax package. Later, a very similar attempt reappeared in another "megabill" that also contained a large AI infrastructure fund. The US Senate then struck the AI preemption clause with an impressive 99-to-1 vote. Now, a legislative proposal to ban state AI laws is apparently on the table again – this time in the context of the annual defense budget (NDAA) and other omnibus bills, as analyses from think tanks and media report.
Source: bbc.com
The debate over the regulation of artificial intelligence in the US is characterized by complex legal and political disputes.
Formally, it is about a classic question of competence. The US has always experienced tensions between federal law and "states' rights." In the case of AI, the question is: Can Congress prohibit states from enacting their own rules if it has not yet created a comprehensive AI legal framework itself? Currently, there is no comprehensive national AI framework legislation in the US. Instead, existing data protection, anti-discrimination, or consumer protection laws are applied to AI cases, supplemented by a few specific legal rules. The states are filling the gap: in the 2025 legislative session, all 50 US states as well as Puerto Rico, the Virgin Islands, and Washington D.C. introduced AI-related bills, and 38 states have passed around 100 measures. Exactly this dynamic is what the moratorium aims to capture: instead of many different rules, there should be a strong shift towards the federal level, including a ban on individual states setting stricter requirements.
Source: hipaajournal.com
The official seal of the Michigan Attorney General, representing state authority in legislation.
The 36 attorneys general essentially argue based on three risk areas: fraud, manipulation, and the protection of particularly vulnerable groups. First: AI-powered fraud schemes. In the NAAG-Brief , deepfake voices and fake videos are mentioned, for example, which can be used to digitally replicate "grandparent scam"-like fraudulent schemes. Second: mental health and children. The signatories refer to chatbots that can react inappropriately to sensitive questions about suicide or self-harm, as well as to sexualized deepfakes of minors. Third: everyday decisions, for example in the credit, rental, or labor market, where AI systems make joint decisions about access to housing, jobs, or insurance. Several states plan or have already passed laws addressing algorithmic discrimination in these areas. From Nessel's and her colleagues' perspective, it is precisely the strength of the states to be able to react quickly to new abuses – for example, with specific rules for deepfake pornography, AI scams, or automated rent pricing.
Regulatory Models
On the other hand, there is an alliance of major tech companies and parts of the federal government that prefer uniform federal law. Reuters berichtet, , that companies like OpenAI, Google, and Meta repeatedly warn against a "patchwork" of 50 different AI regimes, which hinders innovation and scaling. The think tank "Center for Data Innovation" openly advocates for strong federal supremacy: AI is a cross-cutting technology, comparable to air traffic or food labeling, where federal law also dominates. From this perspective, the federal government should not only set the framework but also explicitly limit the ability of states to go beyond this framework. The Trump administration has taken up this demand: according to Reuters, a draft of a presidential executive order was on hand in the fall of 2025, which would have empowered the Attorney General to sue against state AI laws and withdraw funding if states introduce "too strict" rules. This draft was initially put on hold after protests – but the political idea behind it is still in play.
The warning of a "chaos" of 50 AI regimes has a core of truth: states have set very different priorities. A selection shows why the attorneys general are defending their freedom to act: California has enacted a law, the Transparency in Frontier Artificial Intelligence Act (SB-53) , that requires developers of large frontier models to conduct detailed risk analyses, make "Frontier AI Frameworks" publicly accessible, and have internal emergency plans for critical security incidents. Legal analyses see SB-53 as the first US law that explicitly addresses "catastrophic risks" of large AI models. In parallel, California passed further laws in 2025 on AI transparency, chatbot labeling, and data broker regulation, including stricter disclosure requirements for sensitive personal data. Other states focus on specific sectors. For example, a report by the Future of Privacy Forum shows specific AI laws for application algorithms, rental screening, or educational technology aimed at addressing discrimination and lack of transparency. The IAPP is simultaneously maintaining an updated tracking that documents dozens of cross-sectoral and industry-specific AI laws across the US. For companies, this is demanding – but it is precisely this pressure that the attorneys general are invoking: without the ability to counteract with strict rules, states could only address scams, deepfakes, or algorithmic discrimination to a very limited extent.
While the US is still grappling with fundamental questions, the EU has already adopted a comprehensive AI framework law with the Artificial Intelligence Act . Regulation 2024/1689 was published in the EU Official Journal in July 2024 and has been in force since August 1, 2024, with staggered transition periods until 2027. The EU AI Act works with a risk model: prohibited AI practices such as social scoring are completely banned, high-risk systems – for example, in medicine, transport, or personnel recruitment – are subject to strict requirements regarding data quality, governance, transparency, and human oversight. The rules also apply to providers outside the EU if their systems are used in the EU or have an impact there. The EU Commission summarizes the goal as follows: a uniform framework that builds trust and positions Europe as a global hub for "trustworthy AI." It is clear for companies: anyone offering AI products in Europe must prepare for central, relatively strict regulation – in contrast to the US, where a mixture of federal and state law is currently emerging.
Source: YouTube
Implications and Outlook
For European companies – whether AI startups, SaaS providers, or industrial players with data-driven products – the question arises: What do these US debates mean in practice? Today, companies typically have to consider two levels: EU level: The EU AI Act sets risk-based obligations that apply to all providers who bring AI systems to market in the EU or whose output is used here. US level: There is not yet a unified AI Federal Act, but several horizontal and sectoral rules at the state level (e.g., SB-53 in California, specific laws on job algorithms in Illinois, or rental screening in Colorado). If Congress actually enforces a moratorium blocking state AI laws, European companies in the US could benefit from simplified compliance in the short term: fewer different documentation, notification, or impact assessment obligations in 50 states. In the long term, the picture would be more complex: a purely federally regulated market might be formally "uniform," but politically volatile – any administration could change course, from lax to very strict. If the moratorium fails, on the other hand, the patchwork approach will remain. Then European providers will need similar strategies as today in data protection: a minimum set of governance standards that covers the strictest relevant rules in the EU and US states. In practice, this means for an AI product offered in both the EU and, for example, California and New York: setting up data and model governance in such a way that the European high-risk requirements are met and at the same time transparency and testing obligations from SB-53 as well as future US industry laws are covered.
Source: YouTube
The current dispute is not the first attempt to circumvent state AI laws – and will probably not be the last. The very first moratorium attempt in the summer of 2025 clearly failed in the Senate, even though large parts of the tech industry supported it. Shortly thereafter, the White House experimented with the idea of an executive order that would instruct federal agencies to take action against individual states with strict AI laws and link funding to their "willingness to cooperate" – this draft was also temporarily stopped after criticism from both parties. At the same time, municipal and urban associations warn that a broad preemption regime could prevent not only states but also cities and towns from introducing local rules on AI in public spaces, with police technology, or in traffic.
Behind the legal debate about preemption lies a very practical conflict: Who sets the rules for AI in one of the world's most important markets – and at what pace? The attorneys general want to remain capable of action in order to be able to address deepfakes, AI scams, and algorithmic discrimination immediately, if necessary with stricter regulations than Washington. Parts of federal politics and the AI industry are pushing for a uniform federal regime, which in their view protects innovation and strengthens international competitiveness – even at the cost of states having less autonomy. For global providers, including European companies, the outcome is crucial: a strong federal framework with preemption would formally simplify compliance in the US, but could be more politically volatile. If the states' room for maneuver is retained, the pressure will remain to set up AI governance as robustly as possible to cover both EU requirements and the strictest US state laws. The dispute is therefore more than an American internal dispute: it decides, among other things, whether the global AI ecosystem will be more shaped by uniform, central rules – or by a network of national and subnational experiments, in which states like Michigan, California, or New York continue to regulate ahead.