There’s a ton of noise and bad information out there about global AI ethics, especially when it comes to regulating large language models (LLMs) and how international law fits in. If your organization is operating in the AI space, you’ve got to understand how these different frameworks work, or you’re going to run into trouble.
Key Takeaways
- The EU’s AI Act, which went live in 2024, sets up a risk-based system for regulating AI, and it has specific rules targeting general-purpose models like LLMs.
- China rolled out its generative AI rules in 2023, and they demand strict content moderation and data security for LLMs, a very different model of control.
- There isn’t a single global AI treaty. Instead, groups like UNESCO are putting out recommendations that countries are using to shape their own laws, which is creating some overlap on core ideas.
- You can’t have a one-size-fits-all compliance plan. Your company needs a strategy that can handle the different rules on data privacy, bias, and transparency you’ll find in each region.
- The U.S. is taking a different path than the EU. Instead of one big law, it’s using executive orders and agency guidance to create more voluntary, sector-by-sector rules for AI.
“A nuance-free victory for publishers, it fears, could let copyright holders “gatekeep” even innovative and far-reaching uses of their work. On the other hand, a victory for AI companies could constitute “a green light to snub licensing terms and fragment the flow of knowledge between human authors and audiences.””
Myth 1: A Single, Unified Global AI Ethics Law is Imminent
The idea that one global law for AI ethics is coming soon is a complete fantasy. The reality on the ground is a messy patchwork of national and regional rules, each built on different philosophies and enforced differently. While everyone agrees ethical AI is a good goal, the paths countries are taking to get there are wildly divergent. For example, the European Union’s AI Act, which entered into force in 2024, uses a risk-based approach. It sorts AI systems into unacceptable, high-risk, limited-risk, and minimal-risk buckets. If you’re building a high-risk system, think critical infrastructure, hiring, or law enforcement, you’re facing a mountain of requirements on data quality, human oversight, transparency, and cybersecurity. This law is a massive regulatory move that also includes specific rules for general-purpose AI models (GPAI), the category that LLMs fall into. The United States, in contrast, is using existing laws, executive orders, and voluntary frameworks. President Biden’s October 2023 Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence directed federal agencies to develop standards, with the National Institute of Standards and Technology (NIST) told to create evaluation resources while other agencies address AI risks in their respective domains. This is a series of coordinated government efforts, not one single law. A 2025 report from the Brookings Institution put it plainly, noting “the EU’s prescriptive, ex-ante regulation contrasts sharply with the US’s more agile, ex-post enforcement model” (Brookings Institution, “Global AI Governance: Diverging Paths,” 2025). Expecting one global law ignores these deep-seated differences in how governments think about regulation.
Myth 2: LLM Regulation is Primarily About Content Censorship
It’s easy to assume that regulating LLMs is all about controlling the information they spit out, but that’s a narrow and incomplete picture. While content moderation is definitely a piece of the puzzle, especially in some countries, the real regulatory action on large language models is much broader, digging into data privacy, bias mitigation, transparency, and accountability. Take China’s generative AI regulations, which the Cyberspace Administration of China (CAC) put into effect in 2023. Yes, they require generated content to “reflect socialist core values” and ban anything that threatens the state. But the rules also slap strict data security mandates on providers, requiring user consent for data collection and protection for their information. Providers are also on the hook for preventing discriminatory content and must clearly label anything generated by AI. It’s a multi-pronged strategy where content control is just one part. In the EU, the AI Act’s rules for GPAI models are all about risk management. Providers have to run model evaluations, figure out and reduce systemic risks, and get their data governance in order. The focus is on the safety of the model’s design itself, like requiring documentation of the training data to tackle data bias at the source. This is a different world from just policing the output. Even the United Nations Educational, Scientific and Cultural Organization (UNESCO), in its 2021 Recommendation on the Ethics of Artificial Intelligence, pushes for principles like safety and non-discrimination, showing that the global conversation on LLM regulation is about protecting basic rights and guiding responsible development across society (UNESCO, “Recommendation on the Ethics of Artificial Intelligence,” 2021).
Myth 3: International Law Has No Bearing on AI Ethics
Anyone arguing that traditional international law doesn’t apply to AI ethics is missing the point. That view badly underestimates how these legal frameworks are already shaping AI governance. While there’s no single binding treaty for AI, existing principles of international human rights law and new “soft law” instruments are building the foundation for it. Think about it: if an AI system discriminates in hiring, violates privacy, or suppresses free expression, it’s directly running afoul of a state’s existing human rights obligations. The Office of the United Nations High Commissioner for Human Rights (OHCHR) constantly emphasizes this, and their 2024 report on the topic detailed how poorly governed AI can deepen inequalities and create new rights violations. At the same time, international organizations are developing non-binding instruments that often become the blueprints for national laws. We’ve already mentioned UNESCO’s Recommendation on the Ethics of Artificial Intelligence, it’s not legally binding, but it provides a globally recognized set of principles (transparency, accountability, human oversight) that countries are adopting for their own AI strategies. The Organisation for Economic Co-operation and Development (OECD) did something similar with its 2019 Principles on Artificial Intelligence, which have been endorsed by over 40 countries. These “soft law” documents have real teeth in practice, steering policy and creating a common language for ethical AI. To say international law has no role here is to ignore the entire collaborative effort that’s defining the future of AI.
Myth 4: Compliance with One Nation’s AI Laws Guarantees Global Compliance
Assuming that following the rules in one major market like the EU or US means you’re compliant everywhere is a huge, and potentially company-ending, mistake. The fractured state of global AI rules means you absolutely must have a multi-jurisdictional compliance strategy. What’s perfectly fine in one country could get you hit with massive fines and public backlash in another. Let’s look at biometric data processing. The EU’s GDPR already has tight restrictions on using sensitive personal data, and the AI Act goes even further by classifying things like real-time remote biometric ID in public as high-risk, burying it under strict conditions. Other countries might have much looser rules or even define biometric data differently. A 2025 report from the International Association of Privacy Professionals (IAPP) confirmed this, stating that “a ‘one-size-fits-all’ approach to AI governance is untenable given the global variations in data protection laws and cultural norms around privacy” (IAPP, “Global Data Privacy Trends 2025,” 2025). The requirements for algorithmic transparency and explainability also vary wildly. The EU AI Act demands intense documentation and human oversight for high-risk systems. An LLM built to that standard for the EU might still not be good enough for the unique requirements emerging in Japan or Canada. Companies have to spend money on regulatory scanning and localized legal counsel to understand their obligations in every single market, then adapt their models and strategies to fit. Not doing this isn’t just an oversight. It’s a surefire way to get fined into oblivion and lose all public trust. You have to be proactive, informed, and ready to adapt if you want to work with AI globally.
What is the European Union’s AI Act?
It’s a landmark regulation from 2024 that establishes a complete, risk-based legal framework for artificial intelligence. The Act sorts AI systems by their potential for harm and applies stricter rules to higher-risk applications, with specific new provisions for general-purpose AI models like LLMs.
How do China’s AI regulations differ from the EU’s?
China’s 2023 generative AI regulations are heavily focused on content moderation to fit national values, but they also have strict rules for data security and preventing discriminatory outputs. The EU’s AI Act is less about specific content and more about a broad, risk-based approach that stresses data quality, human oversight, transparency, and cybersecurity for high-risk AI.
Is there a global treaty for AI ethics?
No, there isn’t a single, legally binding global treaty just for AI ethics. What we have are influential non-binding recommendations from international bodies like UNESCO and the OECD. These principles are guiding countries as they develop their own policies, creating some international agreement on what ethical AI should look like.
What does “multi-jurisdictional compliance” mean for AI?
It means your organization has to understand and follow the specific AI laws and ethical guidelines for every single country or region where you build or use AI systems. You have to do this because the rules around data privacy, bias, transparency, and accountability are completely different from one place to the next.
How does international human rights law apply to AI?
It applies directly by holding states accountable for ensuring AI systems don’t violate fundamental rights like privacy, non-discrimination, and freedom of expression. Groups like the OHCHR push to apply these existing human rights standards to AI development and deployment to stop abuses and protect human dignity.
