A proposal for legislation has been presented by Anthropic, an AI machine learning company, to regulate the safety of artificial intelligence (AI). The idea is to create a control system that will block the release of AI models if an imminent event threatens human life or causes significant economic damage. The problem is defined as 'imminent' if the model creates or distributes a bio-weapon, surpasses its developer's limitations, or suffers a cyber attack.

The aim is to ensure immediate notification and controllable effects of these events. The company claims this proposal is based on advanced strategies of AI called 'Advanced AI Framework', which began in June. This project included testing AI models and having them evaluated by independent entities.

The article cites three people with privileged information about Anthropic's research who have reported the proposal to US government authorities. These individuals were involved in this process for about two months, during which the company tried to convince potential collaborators to participate in the discussion. The article concludes that the company has presented a legislative proposal that should have been considered by the US government, but it was not accepted.

The organization stated that they expect a positive response from the government, but did not provide further details on verification procedures or possible changes to laws. The document also presents concrete promises such as mandating AI producers to test their models and make results public, and allowing independent entities to evaluate these models. Even though the organization previously released an 'Advanced AI Framework', this time it was presented as a legislative proposal.