Should the development of artificial intelligence be regulated? While 400 scientists have signed an open letter calling for the creation of an international body to oversee AI safety, the American giants of the sector are also calling for regulation. A position that is not without ulterior motives, according to The AI / news analysis.
The fact that companies like Anthropic, creator of the Claude artificial intelligence, and OpenAI, which makes ChatGPT, are calling for regulation may be surprising. However, the establishment of a regulatory body could favor the model chosen by these American commercial companies.
‘Closed’ Models and ‘Open’ Models
Anthropic and OpenAI have opted for so-called ‘closed’ AIs. Their users can use them, but do not have access to their parameters, called ‘weights’. The image used by The AI / news is that of a car for which one has the keys, but cannot lift the hood. It remains possible to test this type of AI, verify that it complies with safety standards, and thus grant it a form of certification.
Conversely, Chinese models, like DeepSeek, are ‘open’. A user can download them onto their own servers and modify their weights to specialize them for a given task. This choice, which allows each developer to adapt the technology as they see fit, would likely be a good way for China to disseminate its technology.
An International Regulation Difficult to Apply
This openness, however, complicates the prospect of international regulation. Once an open model has been disseminated, its user can remove the safeguards intended to prevent misuse or harmful uses, for example to turn it into a tool for cyberattacks. In this context, it is difficult to grant it certification.
This is why the calls for regulation made by Anthropic and OpenAI can also be interpreted as a tactical maneuver. Regulation favoring closed AIs could indeed strengthen their business model against open AIs. A challenge that is not without consequence in the rivalry between the United States and China in the field of artificial intelligence.
