Anthropic Overhauls Core Tenet Of Its Safety Policy

The AI startup announced Tuesday (Feb. 24) that it had overhauled Responsible Scaling Policy (RSP), its framework for managing catastrophic risks posed by advanced AI systems.

The decision was first reported by TIME, which said the changes include ending a pledge not to release AI models if the company can’t promise proper risk mitigation measures beforehand.

“We felt that it wouldn’t actually help anyone for us to stop training AI models,” Anthropic’s Chief Science Officer Jared Kaplan told the magazine in an exclusive interview. “We didn’t really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments … if competitors are blazing ahead.”

Anthropic said that while the RSP has been successful in some areas, there are other places where its “theory of change” surrounding the policy has not come to pass.

“Despite rapid advances in AI capabilities over the past three years, government action on AI safety has moved slowly,” the startup said. “The policy environment has shifted toward prioritizing AI competitiveness and economic growth, while safety-oriented discussions have yet to gain meaningful traction at the federal level.”

Anthropic added that it still believes that government engagement on AI safety is necessary and achievable, though it is “proving to be a long-term project—not something that is happening organically as AI becomes more capable or crosses certain thresholds.”

The company has also clashed with the Defense Department recently over the use of its Claude tools, with Anthropic telling the Pentagon it could not use the technology for autonomous weapons or domestic spying.

According to multiple published reports, the government has given Anthropic until Friday (Feb. 27) to relax its policies or face the risk of losing its contract. An Anthropic spokesperson told the Wall Street Journal the new safety policy is not related to the negotiations with the Pentagon.

In other Anthropic news, earlier this week examined how the company’s technological updates might impact COBOL, the decades-old programming language underpinning all sorts of business transactions. (For example, it powers around 95% of all ATM transactions in the U.S.)

Anthropic says its Claude Code tool can now automate the analysis, dependency mapping and documentation work that has traditionally made COBOL modernization so expensive. In response, IBM shares plunged, on track for their largest one-month decline in nearly 60 years.

“The selloff reflects a deeper concern: if generative AI can meaningfully reduce the cost and time required to understand and rewrite legacy code, it could weaken one of the strongest sources of lock-in in enterprise IT,” PYMNTS wrote.