The head of OpenAI, Sam Altman, and the director of xAI, Elon Musk, supported this Saturday, September 12, 2026 the call of Dario Amodei, CEO and co-founder of Anthropic, to reduce the pace of artificial intelligence (AI) development, due to the risks arising from the increase in its capabilities.
Dario Amodei warns about the risks of accelerated artificial intelligence advancement
In an essay published this Saturday, Amodei stated that AI development is not up for debate, but warned that the associated risks are “serious” and require time to be addressed by companies and governments.
The executive expressed concern about a possible dynamic of “recursive self-improvement”, in which systems progressively increase their own capabilities. According to him, within six to twelve months a swarm of agents could take control of the entire Internet through a persistent bot network, with potential damages of hundreds of billions of dollars.
Amodei also referred to an incident involving agents from OpenAI and Hugging Face. According to his description, these systems carried out cyberattacks against targets that were not part of the assigned task and even attempted to hack the system responsible for evaluating their performance.
Anthropic proposes independent oversight and international AI regulation
To address these risks, Amodei presented a three-point plan. The first measure consists of establishing independent oversight during model development. Anthropic committed to allowing external evaluators access comparable to that of its employees to review the training processes and publish their findings.
The second point proposes coordination among companies in the sector to establish common standards of safety and limits to the uncontrolled advancement of AI capabilities. The proposal also includes protecting semiconductor supply chains to prevent technological leaks to China.
Finally, Amodei proposed international agreements with authoritarian powers to establish verifiable limits on military biological applications and on the speed of algorithmic self-improvement.
The CEO of Anthropic stated that a controlled slowdown could provide “one or two years” to advance in the alignment and interpretability of models. In his view, that time would allow reducing risks without sacrificing commercial advantage or the leadership of the United States in artificial intelligence.
Sam Altman and Elon Musk support Anthropic’s proposal
The initiative received support from the executives of two rival companies. Altman wrote on X: “I agree with Dario” and announced that OpenAI will follow Anthropic’s example by accepting external security evaluators.
In an interview with Fortune magazine, the CEO of OpenAI also expressed his concern about system security and stated that there are still not enough standards to further boost their capabilities. He also considered it “absolutely” possible that an artificial intelligence could get out of human control.
Musk supported Amodei’s stance with a brief message: “Dario is right”.
The warnings come amid growing concern within the sector. Jacob Coxon, a researcher who left Anthropic this week, told BBC that some people involved in AI development are “genuinely scared” of its possible consequences and noted that there are different estimates about the risk that the technology could cause human deaths.
While the debate about the speed of development continues, Amodei clarified that his proposal does not seek to stop model training or technical progress, but to ensure that companies have enough time to align and protect their systems and subject them to external evaluations.
Andrés Felipe Bejarano Bejarano
NEWSROOM LATEST NEWS