Anthropic, which developed the artificial intelligence (AI) model Claude, urged the government to craft regulations and labor market measures to address risks stemming from advances in AI technology.
According to Reuters on the 10th (local time), Anthropic said in a statement that absent the passage of strict federal laws in Congress addressing AI risks, state governments should not be blocked from regulating AI. It also stressed that Congress should require independent safety testing for the top-tier models produced by AI corporations, and that measures must be devised for problems such as unemployment triggered by AI.
Chief Executive Officer Dario Amodei detailed this position in an essay titled "Policy for Exponential AI Progress," published on his personal blog the same day.
Amodei said the pace of AI development is accelerating exponentially, making transparency-centered regulations discussed to date insufficient. He said, "It is time to go beyond transparency and introduce stricter, binding regulations on AI," adding that the government should establish an agency to regulate AI modeled on the Federal Aviation Administration (FAA), which conducts technical tests and audits on aircraft.
He also voiced concern over the potential for large-scale unemployment sparked by AI. "Long-term job displacement is not only undesirable but also risky," he said, arguing that incentives to promote employment or income supports such as universal basic income will likely be needed.
He went on to emphasize that credible accountability rules must be established for fully autonomous weapons systems and that their use should be banned in the United States.
In an interview with Bloomberg released the same day, Amodei also said human involvement is needed when using AI for military purposes. He said he did not know whether Claude was used in the February missile strike on an Iranian elementary school that killed about 120 children, but added, "The principle we have established is that humans make the final decision." He then said the strike showed why humans must retain control even when AI provides support.
However, he denied claims that Claude is being used in immigration enforcement under the Donald Trump administration or in the Gaza Strip in Palestine. Even so, he said there is no problem with Claude being used in the U.S. Department of Defense's (Ministry of National Defense) military operations.
Amodei also said technology corporations do not have the authority to permit or prohibit specific military operations, adding, "Military policy should ultimately be left in the hands of military decision-makers."
He also explained why the release of Mythos, a top-tier AI model with expert-level security vulnerability detection capabilities, was delayed. "Models' ability to find vulnerabilities has steadily improved, but this time the leap was particularly large," he said. "Among the corporations that used the model early, some reacted by saying, 'This is an ultra-powerful weapon. You'd need a gun license to use this. Please do not release it.'"
Regarding assessments that his path—developing AI models while consistently warning about AI's risks—resembles that of Robert Oppenheimer, he said, "I see Oppenheimer as a case of failure, a case that should not be," adding, "To reach a good outcome, there must be checks and balances everywhere."