A new report from the Center for AI Safety just dropped a bombshell: 70% of AI experts think there’s at least a 10% chance that AI will lead to human extinction. That figure alone should be enough to get everyone serious about strong international AI governance and building real content trust to prevent the worst-case scenarios. With stakes this high, we absolutely need global cooperation to have any confidence in AI-generated answers.
Key Takeaways
- The EU’s AI Act, which takes full effect by 2026, creates a risk-based legal framework that sorts AI systems into unacceptable, high, limited, and minimal risk categories.
- A huge global regulatory gap exists, with only 15% of countries having passed complete AI-specific laws, which complicates any effort at consistent international AI governance.
- The U.S. National Institute of Standards and Technology (NIST) published its voluntary AI Risk Management Framework in 2023 to guide organizations in managing AI risks.
- Deep geopolitical divides and conflicting national priorities are blocking the creation of a single, legally binding international treaty on AI.
- A more practical path forward involves a distributed governance model built on interoperable standards and mutual recognition of certifications to build cross-border trust in AI content.
Only 15% of Nations Have Complete AI Legislation
The fact that 85% of the world is a regulatory Wild West for AI is the single biggest problem we face. According to data from the OECD.AI Policy Observatory, as of mid-2026, a tiny 15% of countries have actually enacted complete, dedicated legislation for artificial intelligence. This massive legal gap makes consistent international AI governance nearly impossible. How are we supposed to build content trust in AI outputs when the rules for creating them are wildly different from one country to the next, or simply don’t exist at all?
This absence of a universal legal baseline has a predictable outcome: companies shop for “AI havens,” gravitating to jurisdictions with the weakest rules to save money and move faster. It’s a race to the bottom where ethical considerations and safeguards get thrown out the window for speed. Without common ground, any trust we build in an AI’s answers will be purely local and easily broken. A framework hammered out in Brussels is meaningless in a country with no AI laws, especially when the models themselves are trained on global data and deployed across borders. We are seeing fundamental differences in what is even considered permissible.
The EU AI Act’s Risk-Based Approach
The European Union’s AI Act which will be fully applicable by early 2026, is the first serious attempt at a complete rulebook. It uses a tiered, risk-based approach to regulation, sorting systems into four levels: unacceptable risk, high risk, limited risk, and minimal risk. Anything deemed an “unacceptable risk,” like government-run social scoring or most real-time biometric identification in public spaces for law enforcement, is banned outright. High-risk systems, those used in critical infrastructure, employment, education, and law enforcement, are saddled with a mountain of requirements for data quality, human oversight, transparency, and cybersecurity. These are legal mandates.
For example, an AI system used to assess creditworthiness (a high-risk application) must pass a conformity assessment before it can be sold. Developers have to implement strong risk management systems, prove their data governance is sound, and give users clear information. While critics complain the Act will stifle innovation with its heavy compliance burden, I think it’s the only way to embed accountability and build real user confidence. The EU is setting a precedent that some applications are just too dangerous to deploy without significant safeguards, providing a working model for how international AI governance can start to build shared standards and create content trust.
U.S. NIST’s Voluntary AI Risk Management Framework
In a sharp contrast to the EU’s legal mandates, the United States released its AI Risk Management Framework (AI RMF 1.0) through the National Institute of Standards and Technology (NIST) in January 2023. This is a completely voluntary set of guidelines for organizations to manage AI-related risks throughout a system’s lifecycle, structured around four functions: Govern, Map, Measure, and Manage. The “Govern” function is about creating an internal culture of responsibility, “Map” is about identifying risks, “Measure” is about developing ways to evaluate them, and “Manage” covers how to actually respond to and handle those risks.
Even though it’s voluntary, the NIST framework is getting a lot of traction, with U.S. federal agencies and major tech players like IBM and Microsoft integrating its ideas into their workflows to show they’re being responsible. For example, a company developing an AI for medical imaging might use the NIST framework to systematically document data provenance, record the biases they found during training, and define the human-in-the-loop protocols for doctors who review the AI’s diagnoses. Its widespread adoption could establish a de facto standard, much like other NIST cybersecurity frameworks have, offering a bottom-up path for international AI governance. This approach builds content trust by demonstrating a commitment to responsible development.
The Geopolitical Divide: A Barrier to Unified Global Governance
Even with the clear need for harmonized international AI governance, huge geopolitical divides are stopping a unified global treaty in its tracks. The world’s major powers have fundamentally different philosophies on AI development and the role of the state. China has aggressively pushed state control and surveillance, demonstrated by its massive use of facial recognition and its 2022 “Algorithmic Recommendation Management Provisions” which legally require algorithms to uphold socialist core values. That’s a world away from the human-rights-focused approach coming out of the EU.
These conflicting national interests make a single, legally binding international AI agreement nearly impossible right now. The UN Secretary-General’s High-Level Advisory Body on Artificial Intelligence, set up in late 2023, is trying to find common ground for cooperation, but its proposals will be non-binding. We’ve seen this dynamic before in sectors like climate change, where national priorities kill global consensus. The idea that a single international body could dictate AI policy to every nation, given how strategically important AI is, is just wishful thinking. To build global content trust in AI systems, we have to accept these political realities and look for more practical solutions than some grand treaty.
The Conventional Wisdom: A Single Global AI Treaty is the Answer (and why it’s wrong)
The prevailing view among many policymakers is that a single, global treaty, like the Nuclear Non-Proliferation Treaty, is the only real fix for international AI governance. The logic is simple: AI is a global technology, so it needs a global rulebook. While the sentiment makes sense, this idea is flawed and completely unrealistic given the current geopolitical climate and the sheer speed of AI development. Trying to negotiate one all-encompassing treaty would be an act of futility, burning years only to produce a watered-down agreement that’s obsolete before it’s even signed.
Such a treaty would also never be able to account for the huge differences in ethical and cultural values between nations. What one society considers “responsible AI” might look very different somewhere else. A more effective strategy for building content trust is a distributed, interoperable governance model. This means focusing on bilateral and multilateral agreements on specific AI issues, promoting mutual recognition of national certifications, and developing common technical standards for things like safety and transparency. It’s a network of treaties and agreements, not a single global constitution. This approach allows for the flexibility and adaptation that are absolutely required for a technology as dynamic as AI, aiming for sufficient alignment to ensure we can actually trust these systems.
The path to building international trust in AI’s answers is messy, requiring a real-world understanding of regulatory gaps, technical limits, and geopolitical friction. By focusing on interoperable standards and fostering an ecosystem of mutual recognition, we can gradually build a strong, practical framework for international AI governance that promotes safety and transparency across borders.
What is the primary goal of international AI governance?
The main goal is to create shared rules and standards so that AI is developed and used safely and ethically everywhere. This is the only way people will ever trust what it produces in the long run.
How does the EU AI Act differ from the U.S. NIST AI Risk Management Framework?
The EU AI Act is law, it sets hard rules and penalties for high-risk AI systems. In contrast, the U.S. NIST AI Risk Management Framework is a voluntary playbook of best practices for organizations to follow. You can’t get sued for ignoring it.
Why is a single global AI treaty considered unrealistic by some experts?
A single global treaty is seen as unrealistic because the major global powers (like the U.S., EU, and China) can’t agree on the fundamental rules. The technology also moves so fast that any treaty would likely be outdated before it was even signed.
What does “content trust” mean in the context of AI governance?
In this context, “content trust” is the confidence that users and society have that the information and decisions coming from an AI are accurate, fair, and reliable. Good governance builds this trust by making AI systems accountable and their processes transparent.
What are some practical approaches to improving international AI governance without a single treaty?
Practical approaches include making smaller deals between countries on specific AI topics, agreeing that one country’s safety certification is good enough for another (mutual recognition), and creating common technical standards for things like bias detection and safety testing.
