How Governments Address AI Safety

An Overview of Domestic Regulations and International Frameworks

Report Image

Our report shows how governments address AI safety at the domestic and international level.

Authors

Tommaso Giardini, Nils Deeg

Date Published

16 Nov 2026

Related Topics:

In 2026, AI safety entered the headlines. Reports of AI intruding into external systems started with the OpenAI/Hugging Face incident and were followed promptly by similar disclosures from Anthropic, Google, and Meta. These incidents, alongside a wave of cyber incidents across Europe, raised political attention on AI safety. Moreover, both Anthropic and OpenAI cited cybersecurity as justification for delaying the release of advanced models, Mythos and Astra. 

AI systems can pose risks to people, property and critical infrastructure, whether through malfunction, misuse or adversarial attack. While cyberattacks dominated headlines, governments built safety rules that span a system’s whole lifecycle, including design and testing requirements during development, risk management, monitoring, and corrective measures following deployment, and outright prohibitions of certain systems that pose unacceptable risks.

Our report shows how governments address AI safety at the domestic and international level. It is part of our flagship report on The Fabric of AI Governance, which covers ten concerns.