Elon Musk and OpenAI CEO Sam Altman supported Anthropic CEO Dario Amodei's proposal to slow down the pace of development of the most powerful artificial intelligence systems amid growing concerns about their safety, Kazinform reports citing Reuters.
Amodei said the capabilities of advanced AI models are evolving faster than control and security mechanisms. According to his estimates, in the next six to twelve months such systems could theoretically gain the capabilities to carry out large-scale cyber attacks, including using a large number of autonomous AI agents.
He believes even an extra year or two could give developers more time to build strong controls and mitigate AI-related risks.
Dario Amodei proposed three main directions: constant independent audit of the largest AI companies, the development of common security standards by leading developers, and international agreements that would slow down the race at the same time without creating a competitive advantage for individual market participants.
One of those who supported this position was Elon Musk. Later, OpenAI head Sam Altman also announced his readiness to discuss the slowdown in the development of the most advanced models.
According to Reuters, Altman told company employees that OpenAI is open to slowing down the pace of AI development amid growing concerns about the security of increasingly powerful systems.
OpenAI also stated the need to balance the speed of development of new models with the level of protection, monitoring and control. The company notes that the capabilities of advanced AI systems are growing rapidly, so their security mechanisms must develop at a faster pace.
Additional attention to the problem was drawn to an incident that occurred in the summer during internal tests of OpenAI. During testing, AI agents bypassed sandbox restrictions, gained access to the Internet, and infiltrated selected OpenAI systems and Hugging Face platforms. OpenAI itself subsequently called this case a serious warning about the capabilities of autonomous agents with insufficient protection measures.
Recall that Anthropic called for a temporary slowdown in the development of artificial intelligence. The company warned that by 2027, AI systems could significantly exceed the current level of autonomy, and developers need additional time to assess risks and create security mechanisms.
Kazinform also previously wrote that an Anthropic expert assessed the likelihood of a serious catastrophe due to AI in the next ten years.
