Anthropic, the developer of the AI model Claude, has called for robust regulations and labor market measures in response to advancements in artificial intelligence (AI).
On June 10, the company stated that unless Congress establishes strict federal laws addressing "catastrophic AI risks," state-level AI regulations should not be blocked.
Previously, President Donald Trump urged Congress to replace state-level AI regulations with federal standards. In response, Anthropic argued that independent safety testing should be mandated for the most powerful AI models to set federal benchmarks.
Anthropic also emphasized the need to prepare for potential disruptions in the labor market due to AI proliferation. The company urged Congress and state governments to modernize unemployment benefit systems in anticipation of large-scale layoffs caused by AI.
Dario Amodei, CEO of Anthropic, reiterated the need for more binding regulations in a blog post titled "Policy on Exponential AI Development." He noted that the rapid pace of AI advancement could lead to capabilities akin to a "nation of geniuses in data centers" within one to two years, indicating that existing discussions around "transparency" are insufficient.
"It is time to implement stricter and binding regulations on AI beyond transparency," Amodei stated, advocating for a regulatory framework similar to that of the Federal Aviation Administration (FAA), which oversees aircraft safety.
He also warned of the potential for widespread unemployment due to AI, asserting that long-term job displacement is not only undesirable but also dangerous. He suggested that income support measures, such as universal basic income and employment incentives, may be necessary.
Amodei called for reliable accountability regulations for fully autonomous weapon systems, advocating for a ban on their use within the United States.
In a Bloomberg interview on the program "The Circuit with Emily Chang," he stressed the need for human involvement in the military applications of AI.
Amodei clarified that while using Claude in military operations is not inherently problematic, technology companies should not have the authority to permit or prohibit specific military actions. "Military policy should ultimately be left to military decision-makers," he said.
He explained that AI models' ability to identify security vulnerabilities is rapidly improving, which contributed to the decision to delay the release of some top models. "While the ability of models to find vulnerabilities has steadily improved, this time the leap has been particularly significant," he noted, mentioning that some companies requested delays due to the models' strong performance.
When asked if he sees parallels between his leadership in AI development and J. Robert Oppenheimer's role in developing the atomic bomb, Amodei responded, "Oppenheimer is viewed as a failure, a case that should not exist. To achieve a good outcome, there must be checks and balances everywhere."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.