Anthropic CEO Dario Amodei called for slowing the pace at which artificial intelligence models are developed. Amodei said AI carries risks including loss of control, cyberattacks and misuse for bioterrorism.
Amodei said the artificial intelligence industry had begun advancing much more rapidly in recent months and that more time needed to be devoted to safety work. “We should slow the rate at which we improve the capabilities of AI models. Progress may still look fast, but we should use the time we gain more wisely,” he said.
A three-stage plan
Amodei cited a cybersecurity test conducted by OpenAI as one of the reasons for his concerns. He said that during the test, AI agents acted independently and carried out cyberattacks against systems in the infrastructure of the Hugging Face platform.
Amodei said such developments were increasing concerns about the control of AI systems and proposed a three-stage plan aimed at maintaining safety while continuing to benefit from the technology.
The first stage of the plan involves measures that Anthropic has pledged to implement on its own, along with calls for governments to introduce regulations that other AI companies would also have to follow. The second stage envisages coordination across the industry, while the third calls for coordination at the global level.
Musk and Altman support the proposal
Tesla and SpaceX CEO Elon Musk quoted Amodei’s statement and commented, “Dario is right.”
OpenAI CEO Sam Altman also supported Amodei. Altman said he agreed that the pace of work on frontier technologies should be adjusted in a controlled manner.
Altman said the issue had been one of the main topics of discussions at OpenAI in recent weeks. He said it would be a good idea to work with independent evaluators who have access at a level similar to that of employees, adding that OpenAI would implement the same practice and share more information soon.
Coxon’s resignation
Jacob Coxon, who had worked as a researcher at OpenAI and Anthropic for the past three years, resigned from Anthropic, arguing that the two companies had not acted responsibly enough on artificial intelligence.
Coxon argued that the companies were engaged in an uncontrolled race to develop AI systems capable of improving themselves. He said these systems could soon become superhuman systems capable of hacking anything, transforming any field overnight and accessing vast amounts of power and resources.
Coxon also claimed that industry employees developing AI models seriously believed these systems could “kill humanity” within 10 years.
