Anthropic co-founder Jack Clark called for a "brake pedal" to slow or pause the development of artificial intelligence during a BBC Newsnight interview [1].
This proposal highlights a growing concern among industry leaders that AI capabilities are advancing faster than the safety frameworks designed to manage them. If the technology reaches a point of autonomous self-improvement, the window for human intervention may close.
Clark spoke in London on June 4, 2024, and said that the industry needs a mechanism to mitigate risks [1, 2]. He said that AI models are approaching the ability to improve themselves without human oversight [3, 4]. According to Clark, if the industry does not implement a pause on development, there is a risk of losing control of the technology [2].
The urgency of this request is tied to the rapid evolution of coding capabilities. Clark said that AI could soon write 99% of code [5]. He said that this shift makes the value of junior engineers more dubious as the technology automates foundational programming tasks [5].
While some reports characterize the proposal as a call for a global freeze in development, others describe it as a mechanism to slow the pace of progress [2, 4]. Clark represents Anthropic, which is described as the world’s most valuable AI start-up [4].
"We need a brake pedal for AI," Clark said during the interview [1].
“We need a brake pedal for AI.”
The call for a 'brake pedal' reflects a critical tension within the AI industry between competitive acceleration and existential safety. By flagging the risk of autonomous self-improvement, Clark is pointing toward a 'singularity' event where AI could iterate its own code faster than humans can audit it, potentially rendering current alignment techniques obsolete.
