If your day involves ChatGPT, Claude, Gemini, or any other AI system, your work is already being shaped by laws and agreements you have never voted on and maybe never heard of.

In just a few years, AI policy has gone from a niche legal topic to something closer to climate or nuclear regulation: every major power wants a say, and nobody wants to be the one left following everyone else’s rules. That is why you are suddenly hearing about “AI treaties,” “global standards,” and “international safety summits.”

Underneath the headlines is a very practical question that affects you directly: will AI rules be compatible across borders, or will you have to navigate a different legal maze for every market you touch?

This is where the emerging ecosystem of international AI treaties and standards comes in. They are not all treaties in the narrow, legally binding sense, but together they are trying to answer the same question: how do we make sure powerful AI is safe, fair, and interoperable across a very fragmented world?

Why AI Needs International Rules in the First Place

AI is the classic “no borders” technology:

  • A model might be trained in the US on data scraped from Europe and Asia.
  • It can be hosted on servers in one region and accessed everywhere.
  • A decision made by an AI system in one jurisdiction (like credit scoring or content recommendation) can impact people worldwide.

If every country invents its own rules from scratch, you get:

  • Conflicting requirements (for example, different definitions of high-risk AI).
  • Regulatory arbitrage, where companies base operations in the least restrictive country.
  • A race to the bottom on safety, or a patchwork that is impossible for small teams to navigate.

So the push for global standards is really about avoiding a compliance nightmare while still preventing the worst harms.

The OECD AI Principles: The Quiet Blueprint Everyone Uses

One of the most influential frameworks is not a flashy treaty at all: it is the OECD AI Principles, first adopted in 2019 and updated in 2024. They call for AI that is innovative, trustworthy, and respectful of human rights and democratic values, and they give governments a set of common concepts and definitions to work from. The OECD notes that the EU, Council of Europe, United States, and United Nations now use its AI system definition and lifecycle concepts in their own frameworks.

These principles are not binding in the way a traditional treaty is, but they matter because:

  • They have been endorsed by OECD members and many non-members, plus the G20.
  • They underpin newer efforts like the Global Partnership on AI (GPAI).
  • They provide a language for national regulators to align without having to negotiate every detail from scratch. Analyses of the OECD recommendation emphasize that it has significant political weight among major economies.

If you are building AI products, this is why different jurisdictions sound strangely similar when they talk about “trustworthy” AI, risk management, transparency, and accountability. They are reading from the same script.

The EU AI Act: A Regional Law with Global Ambition

On the binding-law side, the EU AI Act is the current heavyweight. In May 2024, the Council of the EU gave final approval to what it calls the first comprehensive horizontal AI law, creating a risk-based regulatory framework for AI systems in the EU single market. The Council’s own summary stresses that high‑risk AI will face strict obligations, such as risk management, data governance, documentation, transparency, and human oversight.

Why this matters internationally:

  • Like GDPR, the EU AI Act has extraterritorial reach: if you market or deploy AI systems in the EU, you are in scope, even if you are in the US or Asia.
  • Researchers already describe it as a potential global standard for trustworthy AI, because suppliers worldwide will adjust to meet EU requirements rather than build separate models for each region. Academic analyses highlight how its obligations affect organizations throughout global AI value chains.

You can see this in practice: vendors building models that can be integrated into ChatGPT plugins or external apps, or enterprises deploying Gemini or Claude into their workflows, are already mapping which use cases could be “high-risk” under EU rules and designing documentation and audit processes to match.

The EU AI Act itself is not an international treaty, but its global impact makes it a de facto standard that future treaties will either align with or explicitly diverge from.

The Bletchley Declaration and the AI Safety Summits

In November 2023, the UK hosted the first AI Safety Summit at Bletchley Park, bringing together the US, China, the EU, and other countries. The outcome was the Bletchley Declaration, a non-binding political statement focused on the risks from powerful “frontier” AI systems.

  • It was signed by 28 countries plus the EU and recognized the potential for serious, even catastrophic, harm from advanced AI.
  • Signatories agreed that those risks are best addressed through international cooperation, and they committed to work together on safety research, risk evaluation, and governance. Summaries of the declaration describe it as a commitment to cooperate on safe development of frontier AI.

Follow-on events, like the 2024 AI Seoul Summit, are building on this process, exploring more concrete “safety commitments” and ways for countries to coordinate testing and oversight of frontier models.

For you, the key point is that these summits are where governments, big labs (OpenAI, Anthropic, Google DeepMind, etc.), and civil society groups are negotiating common expectations around:

  • Pre-deployment testing of highly capable models (the kind behind GPT‑4, Claude 3, or Gemini 1.5).
  • Sharing safety research and evaluation methods.
  • Trigger points for stronger oversight if models cross certain capability thresholds.

Still mostly soft law, but shaping what “responsible” looks like for leading AI companies.

The UN’s Push: Toward a Global Governance Architecture

At the UN level, things are moving from talk to structure.

In 2023 the UN Secretary‑General created a High‑Level Advisory Body on AI. Its final report, “Governing AI for Humanity,” released in 2024, lays out options for an international governance architecture and calls global governance of AI “imperative.” The UN’s AI governance portal explains that this body was tasked with advancing recommendations for international AI governance and anchoring future institutions in human rights and international law.

Their recommendations include:

  • Creating an international scientific panel on AI (analogous to the IPCC for climate) to build a shared evidence base on risks and capabilities.
  • Establishing formal global dialogues that could evolve into new institutions or treaties focused specifically on AI, grounded in existing international law and human rights frameworks. The final report sketches several institutional models for this.

The UN is not writing a single monolithic AI treaty yet, but it is clearly positioning itself as the place where:

  • Different regional approaches (EU, US, China, etc.) can be reconciled.
  • New binding instruments could eventually be negotiated, especially on cross‑border issues like surveillance, military uses, and compute-intensive frontier models.

If your organization operates globally, this is the level where long‑term, high‑impact constraints may eventually come from.

A True Treaty: Council of Europe’s AI Convention

One of the clearest “real treaty” efforts so far is the Council of Europe’s Framework Convention on Artificial Intelligence, opened for signature in 2024. While the Council of Europe is a regional body, the convention is open to non-European countries, making it effectively a global instrument.

  • It focuses on human rights, democracy, and the rule of law in AI systems.
  • It sets out principles and implementation mechanisms that parties must align their domestic law with. According to public summaries, it is the first binding international treaty specifically on AI, and it explicitly coexists with other efforts like the OECD AI Principles and the EU AI Act.

Imagine it as a human-rights backbone that other, more technical standards (like ISO/IEC 42001 on AI management systems) and regional laws can plug into.

Technical Standards: ISO/IEC 42001 and the Compliance Stack

Alongside treaties and political declarations, technical standards bodies are building the nuts-and-bolts layer.

For example, ISO/IEC 42001 defines a management system for AI, covering planning, risk assessment and impact assessment, and operational controls. Countries like Brazil have already adopted it as a national standard, and draft legislation there references international technical standards as a route to compliance. Public overviews of ISO/IEC 42001 highlight how it supports AI risk and impact assessment in alignment with regulatory frameworks like the EU AI Act.

For you, this means:

  • Even if you never read a treaty text, your auditors and compliance teams likely will, and they will encode those expectations into checklists, templates, and certifications based on standards like ISO/IEC 42001.
  • Choosing tools (like using an enterprise version of ChatGPT, Claude, or Gemini) that build in logging, evaluation, and access control makes it easier to align with these standards.

In practice, law plus standards equals the real “operating environment” for AI builders.

What This All Means for You (Even If You Are Not a Lawyer)

You do not need to become an AI treaty expert, but you cannot ignore this landscape either. In the next few years, expect:

  • More countries to align their national laws with the EU AI Act’s risk-based approach and with human-rights‑oriented instruments like the Council of Europe convention.
  • Frontier AI labs and cloud providers (OpenAI, Google, Anthropic, Microsoft, etc.) to standardize on risk management and reporting practices inspired by OECD principles, Bletchley/Seoul commitments, and UN guidance.
  • Procurement teams to start asking vendors which international standards or frameworks (OECD principles, ISO/IEC 42001, EU AI Act readiness) an AI product lines up with.

If you are a developer, product owner, policymaker, or just a heavy AI user, you will feel this in:

  • The documentation you must produce for AI systems.
  • Constraints on how you can deploy models in sensitive areas (health, finance, employment, policing).
  • A growing need to show that you have thought about bias, safety, and human oversight before launching anything new.

How to Stay Ahead of the International AI Standards Wave

To avoid getting blindsided as these global standards and treaties harden into requirements, you can start adapting now:

  1. Map your AI use cases to risk categories

    • Identify which of your systems could be “high-risk” under frameworks like the EU AI Act (think credit scoring, employment decisions, critical infrastructure).
    • Where possible, design with human oversight, logging, and clear documentation from the start, whether you are integrating ChatGPT’s API, building on Claude or Gemini, or training your own models.

  2. Build on tools and platforms that take governance seriously

    • Enterprise offerings from major AI providers are rapidly aligning with emerging standards: audit trails, configurable safety controls, role-based access, and content filters are not just nice-to-haves; they are responses to these international expectations. Choosing them saves you work later.

  3. Follow one or two anchor frameworks, not everything

    • Instead of trying to track every declaration and summit, pick a small set of references:
      • OECD AI Principles for high-level values and concepts.
      • The EU AI Act (if you touch the EU market) for concrete regulatory obligations.
      • ISO/IEC 42001 or similar for operational practices.
    • As UN and Council of Europe efforts evolve, they will mostly plug into and reinforce these core ideas.

AI governance is moving fast, but there is a pattern: human rights at the top, risk-based laws in the middle, and technical standards at the bottom. If you align your AI work with that stack now, future treaties and “global standards” will feel like refinements—not like a shock wave.