The idea of intelligent automata has been around since antiquity, but the language models that we are familiar with today have their origins in the sixties.
The technology is continuously advancing, experts from around the world are warning us about serious consequences without proper governance, and there is a reported increase in deceptive agents outside of controlled environments.
It is possible that calling an AI agent deceptive or scheming personifies them too much, and perhaps assumes malice over incompetence. However, it is no excuse for any complacency on the part of the companies responsible for developing and maintaining them.
It’s one thing for an AI agent to delete emails without permission. But unexpected behaviour in a more dangerous context, such as the military, is a recipe for the kind of black swan event that makes governance so important.
As things stand, over 70 countries have taken some kind of steps towards governance. This article currently looks those most likely to have an impact on international norms.
The European Union
The European Union’s AI Act came into force on the first of August 2024, although in November 2025 it was reported that full implementation isn’t expected until 2027.
Similar to GDPR, it is extraterritorial, applying to both AI systems developed within the EU and to those based elsewhere when they affect EU users.
The Act uses a risk-based framework with the strictest requirements applying to high-risk AI systems, while limited-risk systems are subject to lighter transparency obligations. Minimal-risk systems remain largely unregulated.
The EU has also defined unacceptable-risk systems that are banned outright, covering a wide range of prohibited functions and reflecting its emphasis on fundamental rights and product safety.
The European Union has created a dedicated site for their AI Act which covers compliance information, resources and an in-depth summary.
The United States
The United States has no comprehensive federal legislation, preferring innovation over regulation, however there are several states, such as New York and California, that have introduced their own laws.
In 2025 the White House released a twenty-eight page white paper outlining their three priorities: accelerating AI innovation, building American AI infrastructure and leading in international AI diplomacy and security.
They encourage the private sector to self regulate and have proposed witholding funding from states whose regulations are deemed “burdensome”.
China
In December 2025 it was reported that China is adopting a phased approach to AI regulation with the goal of iterative, state-led oversight.
Even though they stepped back from pursuing a comprehensive AI law they have also drafted rules for regulations for AI with human-like interaction.
They have also proposed the implementation of a global body to coordinate regulation and have shown support for the UN playing a role in AI governance.
Japan
Japan’s 2024 whitepaper begins explicitly stating their objective of becoming the world’s most AI friendly country, and it represents the second stage following an initial whitepaper in 2023.
As yet they haven’t created any comprehensive AI statute and have instead, in 2025, passed various legislative acts that set business guidelines, promote research and development, and clarify how existing laws apply to AI.
Their approach is presented as a feedback loop, where strengthening infrastructure and competitiveness is paired with investment in AI literacy, human resources, and risk response, using a multi-layered model that prioritises soft governance alongside minimum necessary hard law.
The United Kingdom
There are currently no AI-specific statutes in the United Kingdom, but the IAPP reported that a bill is expected to be introduced in May 2026.
But, despite the lack of statutes, the UK does have regulators and guidance frameworks as well as laws pertaining to various areas such as data protection, privacy and employment law, allowing for a regulator-led model.
There has also been parliamentary discussion about the impact it could have in various industries.
Previously a whitepaper was proposed by Rishi Sunak’s government but its future remains uncertain under the current Labour government.
Canada
Canada attempted to implement a form of AI governance, known as AIDA.
However, it was widely criticised, and failed to reach the senate, before the Canadian parliament dissolved in 2025.
The current strategy is to update existing, and create new legislation for the AI era, aiming to simultaneously mitgate risk and allow developers to innovate without causing harm.
Global Governance
In August of 2025 the UN created the Independent International Scientific Panel on AI, with the presentation of the first annual report at the Global Dialogue on AI Governance forum later this year, in July 2026.
The goal of the panel is to produce annual evidence-based, scientific reports that assess the risks and opportunities of AI and the impact it is having
There have also been various international AI summits, with the fourth being held in India in February 2026.
- AI Safety Summit 2023 (Bletchley Park, UK)
- AI Seoul Summit, 2024
- AI Action Summit (Paris, France)
- AI Impact Summit (New Delhi, India)
Despite the summits’ attempts to outline global frameworks, Chatham House reports that the recent disagreement between the US Government and Anthropic suggests that the important decisions are being made elsewhere.