4.2.5 Diffuse and Informal Interventions in AI Development

Governments regularly host AI forums to highlight domestic innovation and AI achievements in a bid to influence global AI model development. ¹⁹ Such forums can also achieve political objectives. For example, the World Artificial Intelligence Conference (WAIC) in Shanghai demonstrated local developers’ resilience and advancement despite U.S. export controls, while Saudi Arabia’s Global AI Summit focused on “harnessing AI for the good of humanity”.

Furthermore, there is evidence that governments have influenced and attempted to influence the actions of AI companies without the use of formal legislation. Examples include the BidenHarris Administration securing voluntary commitments from AI companies, and the Sunak-led UK government securing an agreement for the UK’s AI Safety Institute to test models pre-launch, among others. ^{20 21} While these initiatives have been framed largely in positive, safety/rightsenhancing terms, governments can also use the threat of regulatory action or the withholding of benefits to “jawbone” providers and inappropriately coerce them into changing their models to meet governmental objectives or expectations. Such informal interventions can also include outright threats linked to protectionism, which can impact company behaviour, such as the Trump administration’s trade-related threats to the EU due to the implementation of the Digital Services Act, which may impact the compliance of U.S.-based companies operating in these jurisdictions.

_{19 For example, the AI Impact Summit in India and the AI Action Summit in France}

_{20 Specifically, ensuring the safety and security of AI products before public release, prioritizing cybersecurity, and building public trust through mechanisms like watermarking and transparent reporting on capabilities and risks.}

_{21 Specifically, providing the UK’s AI Safety Institute with prerelease access to OpenAI, Google DeepMind, and Anthropic’s most advanced AI models for safety testing.}