According to Time magazine, the American AI startup Anthropic has achieved great success and bills itself as the most safety-focused company among top AI research laboratories. However, Anthropic executives told the Times,The company is abandoning a core promise of its flagship safety policy.
Anthropic promised in 2023 that it would never train an AI system unless it could ensure adequate security measures in advance. This commitment becomes a core pillar of Anthropic's Responsible Scaling Policy (RSP). For years, Anthropic executives have touted this commitment as proof that they are a responsible company that can resist the temptation of the market and is not in a rush to develop potentially dangerous technology.
But in recent months, the company decided to overhaul RSP.The decision includes abandoning an original commitment not to release AI models without ensuring appropriate risk mitigation measures in advance.
"We believe that stopping training AI models doesn't actually help anyone," Jared Kaplan, chief science officer at Anthropic, told TIME in an exclusive interview. "As AI advances rapidly,We do not feel that it is reasonable for us to make unilateral commitments when our competitors are advancing rapidly."
Anthropic's new policy reviewed by The Times includes: a commitment to be more transparent about AI security risks, including additional disclosure of the performance of Anthropic's own models in security tests; a commitment to invest and exceed the level of competitors in security; and a commitment to "delay" Anthropic's AI development process if leadership believes that Anthropic is the leader in the AI race and believes that the possibility of catastrophic risks is high.
But overall, this adjustment of RSP has greatly reduced the constraints on Anthropic's security policy. Previously, the policy explicitly prohibited companies from training models beyond a certain level without appropriate security measures.
Anthropic currently faces stiff competition from rivals such as OpenAI, Elon Musk's xAI and Google, all of which are releasing cutting-edge tools on a regular basis. The company is also locked in a dispute with the U.S. Department of Defense over the use of its Claude tool. Anthropic has previously told the Pentagon that the tools may not be used for domestic surveillance or lethal autonomous operations. However, the U.S. Department of Defense also issued an ultimatum to Anthropic, threatening to terminate Anthropic's contract if its use was restricted.
Anthropic stated that this adjustment to the security policy was an update based on the speed of AI development and the lack of relevant regulations at the federal level.
An Anthropic spokesperson said that this adjustmentDesigned to help companies compete with multiple competitors in an uneven policy context.In this environment, companies must judge for themselves whether security safeguards are adequate. She also said the security commitment had nothing to do with Pentagon negotiations.
“The policy environment has shifted toward prioritizing AI competitiveness and economic growth, while safety-oriented discussions have yet to gain substantial ground at the federal level,” Anthropic said in a blog post announcing the change. The company said it remains committed to adhering to industry-leading safety standards.