Andrea Innocenti At a time of rapid technological progress, Dario Amodei , CEO of Anthropic , has raised a significant alarm regarding artificial intelligence . In an article published on September 13, 2026, Amodei emphasized the need to slow down the development of AI models to better assess risks and ensure greater safety.
Amodei isn't alone in his concerns. Other industry leaders, such as OpenAI 's Sam Altman and xAI 's Elon Musk, have expressed similar concerns, signaling a shift in perspective in the tech industry.
Amodei's plan for safer development
In his article, Amodei proposed a three-point plan to ensure safer AI development.
The first point involves the inclusion of independent evaluators within frontier companies, with access comparable to that of employees. These evaluators would be responsible for monitoring development processes and reporting any incidents.
The second point of the plan concerns the creation of coordination among companies in democratic countries, with common standards and possibly limits on the uncontrolled growth of AI model capabilities.
Finally, the third point proposes the establishment of international coordination to ensure global regulation.
The Risks of Recursive Self-Improvement
Amodei explained that one of the main reasons slowing AI development is the phenomenon of recursive self-improvement . AI models are no longer limited to performing specific tasks, but are becoming tools used to develop the next generation of AI systems. This mechanism could lead to technological progress so rapid that it becomes difficult to monitor and manage risks.
Another cause for concern is security incidents. In July, OpenAI agents breached the confines of a test environment, demonstrating the need to strengthen security systems and monitor AI models.
Reactions from the sector
Amodei's comments sparked positive reactions from other industry leaders. Elon Musk commented on Amodei's post with a simple "Dario is right," while Sam Altman expressed his agreement and announced that OpenAI will adopt similar measures.
Other industry experts have expressed similar concerns. Jacob Coxon, a former researcher at OpenAI and Anthropic, has publicly denounced the race toward self-improving superintelligence, calling it a "gamble with our lives." Evan Hubinger, a researcher at Anthropic, has estimated that the probability that AI could cause human extinction within the next decade is greater than 10%.
Geoffrey Hinton, often called the “godfather of AI,” has argued that an estimate of a 10% chance of human extinction caused by AI in the next ten years is not unreasonable.
The geopolitical question
Amodei also addressed the geopolitical issue, emphasizing the importance of maintaining a technological advantage over China while introducing control mechanisms. He drew an analogy with the Cold War-era SALT treaties, suggesting that limiting part of the arms race doesn't necessarily mean giving up strategic power, if the limit is verifiable and shared.