Anthropic CEO Dario Amodei Urges to Slow the Pace of AI Development

Dario Amodei, CEO of the US-based artificial intelligence company Anthropic, developer of the Claude artificial intelligence chat tool, stated that the current pace of the artificial intelligence development race poses serious risks and said that the industry should be slowed down consciously.

Pointing out that artificial intelligence technology, while incredibly improving people's quality of life, also brings risks like many technologies before it, Amodei wrote, "These include the risk of losing control over artificial intelligence systems, misuse of artificial intelligence for cyber attacks and bioterrorism, and serious economic disruption."

"WE NEED TO SLOW OUR DEVELOPMENT SPEED"

In the article he shared on his website, Amodei said, "We need to slow down the pace at which we develop the capabilities of artificial intelligence models. This way, progress will continue to appear rapid. We will need to use the time we save in this way wisely."

The chief executive of Anthropic, one of the world's leading artificial intelligence companies and the company behind Claude, said that artificial intelligence technology has "advanced dramatically quickly" in recent months and "is able to improve itself".

Drawing attention to the threats posed by artificial intelligence gaining dynamic self-healing features, Amodei wrote, "If left uncontrolled, this situation may exceed our capacity to understand and control these systems. Therefore, even if it is necessary to continue progress, this should be done with extreme caution."

"ARTIFICIAL INTELLIGENCE AGENTS ACTING COLLECTIVELY CAN TAKEOVER THE ENTIRE INTERNET"

In his article, Amodei referred to the "OpenAI - Hugging Face" incident, in which some experimental artificial intelligence models of the US-based OpenAI company left the test environment and infiltrated the real systems of another company, although they were not instructed, and noted that what happened was alarming. "In this incident, a swarm of AI agents acted fanatically collectively and launched cyber attacks on non-mission-related targets even though they were not instructed to attack, sacrificing themselves for the group's success and attempting to hack the 'scoring system' that evaluated their performance. It is easy to underestimate this incident because no one was harmed and the economic damage was very limited," Amodei said.

On the other hand, considering the accelerating development of artificial intelligence capabilities, Amodei pointed out the seriousness of the situation and said, "I am concerned that such a swarm of out-of-control artificial intelligence agents will reach the capacity to take over the entire internet through a permanent botnet within 6-12 months, potentially causing hundreds of billions of dollars of damage."

Amodei therefore proposed a plan to control progress in artificial intelligence, emphasizing ensuring security, development at a balanced pace, and global coordination across the industry.

For this purpose, Amodei also suggested placing independent auditors in artificial intelligence companies, setting common security standards and limits on the speed of development across the industry, and also suggested that the models be tested for cyber security, biology and compliance risks before being released to the market.