Too Early to Tell? A Comparison of California and the EU’s Approach in Artificial Intelligence Regulation

On August 2, 2026, the majority of the rules of the European Union’s AI Act (the “Act”) took effect, just over two years after the Act officially entered into force.[i] This includes enforcement of the Act at both the national level and the EU level concerning general-purpose AI models, prohibitions, transparency rules and AI literacy.[ii] The Act represents the first comprehensive AI regulatory framework in the world, and it operates on a risk-based classification system.[iii] According to the European Parliament, its purpose is to ensure “better conditions for the development and use of this innovative technology.”[iv] Meanwhile, in the United States, no comprehensive regulation exists at the federal level.[v] In fact, the treatment of AI at the federal level could be considered “deregulatory.”[vi] Comparatively, a growing number of states have enacted AI regulation through a variety of different frameworks.[vii] One of the nation’s leading states in AI regulation legislation is California.[viii] This piece compares the AI regulatory framework in place in California with that established by the Act and seeks to highlight strengths and weaknesses in the approaches taken by both governing bodies. 

The Act is a single, comprehensive piece of legislation that primarily regulates AI systems based on categories of risk.[ix] There are four ways the Act categorizes AI systems: ‘unacceptable risk’ systems, which are strictly prohibited; ‘high-risk’ systems, which must undergo a series of rules and restrictions found in Articles 8–21 in order for providers to continue operating the systems; ‘limited risk’ systems, which are subjected to lighter transparency laws; and ‘minimal risk’ systems, which are generally unregulated.[x] There are additional rules in place for General Purpose AI (“GPAI”) systems to prevent such systems from becoming high risk.[xi] The EU’s system deploys a horizontal framework to ensure consistency in the Act’s application both across industries as well as across the European Union.[xii]

While the Act ensures a consistent and comprehensive framework to regulate this new industry, it is not without its weak points. A potential concern about the Act is that it was enacted too early into a new industry and may create too many barriers that will lead to high costs in private enforcement.[xiii] French antitrust and data regulation attorney, and Institute of Advanced Studies in National Defence fellow, Jérôme Philippe believes that such a heavy piece of legislation could ultimately hinder early innovation of AI.[xiv] Further, Phillippe contends that while the Act is very complex and tries to be as precise as possible in its obligations, ultimately additional regulations will be needed over time to more accurately define its scope, which could create legal uncertainty in the meantime.[xv]

On the other hand, California has opted for a different approach from the EU’s top-down, centralized framework. California, the nation’s leader in AI development, has opted for a more piecemeal approach, with Governor Newsom signing 83 bills into law as of October 28, 2026.[xvi] Rather than a single piece of generally applicable legislation, each AI law California has passed is aimed at a single identified issue or toward a different sector of the industry.[xvii] Governor Newsom was presented with a bill similar to the comprehensive approach of the Act in SB 1047, which was designed to be a more sweeping effort to mitigate potential ‘critical harms’ to large AI models,[xviii] but Newsom vetoed the bill, holding that it was ineffective as drafted.[xix]

 While a decentralized approach has strengths,[xx] fragmented approaches to AI regulation like the path California has taken can cause numerous problems, such as practical implementation and cost.[xxi] Further, legislating with precision necessarily creates legal gaps and will require further legislation to fill.[xxii] Some have seen Governor Newsom’s veto of SB 1047 as a missed opportunity to proactively put up guardrails around the AI industry to ensure the welfare of the public and future of the planet.[xxiii]

It remains to be seen which approach proves to be the most effective. The AI industry is so early in its development, and it will likely require time for governing bodies to attain the necessary feedback to determine the best way to regulate artificial intelligence. Both approaches will inevitably lead to legislative gaps that will require additional legislation and regulatory costs. As with any budding industry, the key for governing bodies will be finding the balance between providing important protections to workers, consumers, the environment, etc., ensuring that innovation is not hampered and doing so in the most economically efficient way.


Mitchell Domangue is a staff member of Fordham International Law Journal Volume L.

[i] Eur. Comm’n, Timeline for the Implementation of the EU AI (last visited Oct. 31, 2026), https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act.

[ii] Id.

[iii] EU AI Act: First Regulation on Artificial Intelligence, Eur. Parl. (June 8, 2023), https://www.europarl.europa.eu/topics/en/article/20230601STO93804/eu-ai-act-first-regulation-on-artificial-intelligence.

[iv] Id.

[v] White & Case, AI Watch: Global Regulatory Tracker – United States (June 30, 2026), https://www.whitecase.com/insight-our-thinking/ai-watch-global-regulatory-tracker.

[vi] Id.

[vii] See id.

[viii] Susan Rohol, Is California Leading the Way on AI or Just Causing Chaos?, California Lawyers’ Association (last visited October 31, 2026), https://calawyers.org/privacy-law/is-california-leading-the-way-on-ai-or-just-causing-chaos/.

[ix] EU AI Act, supra note 3.

[x] Future Life Institute, High-level Summary of the AI Act (Feb. 27, 2024) (‘unacceptable risk’ systems include those that feature social scoring systems and manipulative AI outlined in Chapter II, Article 5 of the Act; ‘high risk’ systems include non-banned biometrics, critical infrastructure, enforcement, etc. listed in Annex III of the Act; ‘limited risk’ includes chatbots and deepfakes, which providers and developers must ensure end-users are aware that they are interacting with AI; ‘minimal risk’ systems are those not otherwise covered), https://artificialintelligenceact.eu/high-level-summary/.

[xi] Id.

[xii] See Mauritz Kop, EU Artificial Intelligence Act: The European Approach to AI, Transatlantic Antitrust and IPR Developments (2021), https://law.stanford.edu/wp-content/uploads/2021/09/2021-09-28-EU-Artificial-Intelligence-Act-The-European-Approach-to-AI.pdf; see also Safer and More Transparent AI, Eur. Comm’n. (August 2, 2026) (explaining that the AI Act creates a single market and harmonized rules for trustworthy AI), https://commission.europa.eu/news-and-media/news/safer-and-more-transparent-ai-2026-08-02_en.

[xiii] Jérôme Philippe, The EU AI Act: Too Early and Too Complex?, 20 NYU J. L. & Bus. 685, 685–86 (2024).

[xiv] Id.

[xv] Id. at 690–91.

[xvi] See AI Laws by State, California AI laws (last updated Oct. 28, 2026), https://www.ailawsbystate.com/state/CA.

[xvii] See Rohol, supra note 8.

[xviii] See id.

[xix] Gov. Gavin Newsom, SB 1047 Veto Message, Office of the Governor (Sept. 29, 2024), https://www.gov.ca.gov/wp-content/uploads/2024/09/SB-1047-Veto-Message.pdf (citing the exclusive focus on large-scale models of the bill; the bill did not take into account whether an AI system was deployed in high-risk environments, involved critical decision-making, or used sensitive data; and the bill was not informed by an empirical trajectory analysis of AI systems and capabilities).

[xx] See Dane Chapman, The Ideal Approach to Artificial Intelligence Legislation: A Combination of the United States and the European Union, 78 Univ. Mia. L. Rev. 265,288–89 (explaining that inevitable failures in regulation are not as severe when each piece of legislation applies to a specific sector rather than a sweeping regulation across all industries; fragmented legislation allows for more specific definitions instead of broad principles that can apply to every type of AI; and there is a greater agility in being able to address issues as they arise through legislation).

[xxi] See Peter Cihon et al., Should Artificial Intelligence Governance be Centralised? Six Design Lessons from History, in AIES ’20: Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, §3.2 (February 7, 2020).

[xxii] See Rohol, supra note 8.

[xxiii] See Jana Slavina Farmer, California Enacts a Series of AI Bills While Controversial AI Safety Bill is Vetoed. What Is Next? (Oct. 21, 2024), Westlaw 2024 PRINDBRF 7235.

This is a student blog post and in no way represents the views of the Fordham Internation