Anthropic and OpenAI adjust their security rules amid accelerating AI competition.

Anthropic has removed a core safety commitment from its responsible scaling policy, no longer promising to pause training until risk mitigation measures are fully in place. Anthropic's Chief Scientific Officer, Jared Kaplan, told TIME that in the context of rapid AI development, unilaterally making a training halt commitment is meaningless if competitors continue to push forward. OpenAI has also revised its mission statement, removing the word "safely" from its 2024 IRS document. Previously stated as building general AI that "safely benefits humanity," it now states "ensuring that general AI benefits all of humanity." Edward Geist, a senior policy researcher at RAND Corporation, stated that the advanced AI envisioned by early AI safety advocates differs fundamentally from current large language models, and this terminology change reflects companies' desire to signal to investors and policymakers that they will not back down in economic competition due to security concerns. Anthropic recently completed a $30 billion funding round, valuing the company at approximately $380 billion; OpenAI is currently pursuing a funding round of up to $100 billion, backed by Amazon, Microsoft, and Nvidia. Meanwhile, Anthropic's refusal to grant the Pentagon full access to Claude has created a public rift with Defense Secretary Pete Hegseth, casting uncertainty over its defense contracts. (Decrypt)