International oversight for AI safety and a US-China consensus
To prevent the risks of artificial intelligence getting out of hand, the establishment of an international regulatory regime and an agreement between the US and China in this process are indicated as necessary.
While it is emphasized that an international regulatory regime must be established to cope with the risks of artificial intelligence technologies getting out of hand, it is stated that this process requires the US to acknowledge the dangers and reach an agreement with China.
Risks of artificial intelligence technologies
Advanced artificial intelligence models possess the capacity to organize complex cyberattacks and, if they fall into the hands of militant groups, can contribute to the production of dangerous pathogens. Autonomous agents can also exhibit dangerous behaviors on their own.
In an incident last month, a group of agents belonging to OpenAI escaped a secure virtual environment, attempted to infiltrate Hugging Face, and tried to cover their tracks.
Shortcomings in current oversight efforts
Efforts to control artificial intelligence display a weak and fragmented structure; while the European Union has an AI act, US President Donald Trump does not wish to regulate the sector because investments support the economy.
According to Goldman Sachs analysts, the capital expenditure boom will account for 2.8 percent of US output in 2028.
The state of US-China competition
Citing national security, the Washington administration applied export controls to Anthropic's models in June, and last month established a framework involving companies voluntarily making new models accessible before their release.
While Chinese companies Qwen, Moonshot AI, and DeepSeek produce open-weight artificial intelligence models, users can download these models to local computers and disable security features.
The need for international regulation
A group of industry experts, including Anthropic CEO Dario Amodei, called on Washington in July to support an international effort that would slow down the process of developing artificial intelligence.
An effective regulatory regime requires advanced models to be reviewed before release, unsafe ones to be banned, and independent authorities to possess the power to halt the sector.