Anthropic called for a coordinated and verifiable global freeze on the development of powerful artificial intelligence systems on June 5, 2024 [1].
The request highlights a growing tension between the speed of AI innovation and the ability of governments to implement safety guardrails. If advanced systems begin to improve themselves without oversight, the industry risks creating technology that cannot be managed or deactivated.
Anthropic, an artificial-intelligence research company, said the pause is necessary to prevent the loss of human control over AI [1, 2]. The company said that rapid advances could produce self-improving AI that outpaces society’s ability to manage risks [2, 3]. This trajectory, according to the company, could potentially lead to devastating cyber-attacks [2, 3].
While the company describes the move as a global freeze [1], other reports characterize the request as a global pause [3]. Regardless of the terminology, the goal is a coordinated effort across AI labs to stop the deployment of increasingly powerful models until safety standards are verified.
The company said that the risk of self-improving models is a primary driver for this call. Such systems could theoretically rewrite their own code to become more efficient or powerful, leaving human developers unable to predict or stop their behavior [2, 3].
“Anthropic called for a coordinated and verifiable global freeze on the development of powerful artificial intelligence systems.”
This call for a freeze represents a significant shift in the AI industry, where a leading developer is now publicly advocating for the cessation of its own primary activity. It suggests that the internal risk assessments at top labs may be diverging from the public-facing narrative of safe, incremental progress. If other major players like OpenAI or Google do not join the freeze, the move may create a fragmented regulatory environment where some firms adhere to safety pauses while others accelerate to maintain a competitive edge.
