Governments pursue divergent responses to expanding AI risks

Governments are adopting different approaches to AI risks as increasingly capable systems spread into cybersecurity, personal relationships and other sensitive areas.

In the United States, President Donald Trump announced a voluntary agreement under which about two dozen technology companies committed to independent safety audits and controls addressing cybersecurity, biological and chemical threats, and unintended model actions. Signers included Anthropic, OpenAI, Nvidia, Meta, Alphabet and SpaceXAI. The accord carries no legal requirements, though it says its recommendations could eventually be codified.

Separately, the Federal Trade Commission is investigating OpenAI, Anthropic and other AI companies over risks associated with their products, Fast Company reported. The inquiry shows that existing consumer-protection authority could be used alongside voluntary commitments without creating AI-specific rules.

China took a more prescriptive approach on July 15, introducing regulations covering AI systems capable of sustained emotional interaction. The rules apply both to dedicated companion apps and to broader language models that imitate human relationships, according to the BBC.

Taiwan, meanwhile, is working with Anthropic, OpenAI, Google and Microsoft to obtain advanced models that can scan government systems and critical infrastructure for vulnerabilities. Digital Affairs Minister Lin Yi-ching said Taiwan is also drafting legislation for agentic AI, while emphasizing rapid detection and defense because technological development may outpace lawmaking.

The policy debate is complicated by open-weight models, which users can run and modify independently. Researchers cited by CBS News said such models can improve access and customization but make centralized monitoring and enforcement of safeguards more difficult.