Anthropic has quietly dismantled one of the most concrete safety commitments in the AI industry. The company's Responsible Scaling Policy, first published in 2023, originally included a binding pledge to pause development of any AI system if it could not guarantee in advance that its safety measures were adequate. That commitment is gone.
According to "Hvylya", citing a TIME investigation, Anthropic rewrote the policy in late February. Co-founder and chief science officer Jared Kaplan said it had been "naive" to think the company could identify clear lines between danger and safety. "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," he said.
The revised policy still includes safety measures. Anthropic committed to greater transparency about how its models fare in safety testing and pledged to match or surpass the safety efforts of competitors. The new version also promises to "delay" development if leaders both consider Anthropic to be ahead in the AI race and believe the risks of catastrophe are significant.
But the overall effect left Anthropic far less constrained by its own safety rules. The original policy had been touted as evidence that the company was willing to withstand market pressure in the sprint for superintelligence. Its removal came in the same week Anthropic was locked in a standoff with the Pentagon over the use of Claude in military applications - a confrontation in which the company maintained it was standing by its values at great cost.
The timing raised uncomfortable questions. Anthropic had built its reputation as the safety-first lab, the one willing to sacrifice speed for caution. Dropping the binding pause while simultaneously claiming the moral high ground against the Pentagon underscored a tension that runs through the company's identity. The company's own chief scientist believes fully automated AI research could be a year away. The braking mechanism has been loosened just as the road ahead grows steeper.
Also read: The Sycophancy Trap: Why Flattering AI Chatbots Won't Become Echo Chambers After All