Anthropic Wants AI to Slow Down—But Can Its Rivals Afford To?

Anthropic Wants AI to Slow Down—But Can Its Rivals Afford To?
The safety debate sharpened this week after former Anthropic researcher Jacob Coxon resigned publicly, accusing leading labs of “gambling with our lives” as they pursue self-improving systems. Other Anthropic staff amplified the alarm, helping push a once-specialist argument about catastrophic AI risk into the wider public and political arena.

On Saturday, Anthropic CEO Dario Amodei turned that anxiety into a concrete proposal: “We must slow the pace at which we improve the capabilities of AI models.” He cited rapidly improving systems, including their growing ability to help build their successors, and a recent rogue-agent cybersecurity episode as reasons to buy time.

Amodei’s three-part plan calls for permanent independent evaluators with employee-level access inside frontier labs; shared safety standards among companies in democratic countries; and limited coordination with authoritarian governments on clear red lines, such as AI-assisted biological weapons. Anthropic says it will immediately take the first step, allowing outside evaluators to publish findings without company editorial control.

The immediate reaction suggested unusual common ground among competitors. OpenAI chief Sam Altman said, “I agree with Dario that we need to pace the frontier,” and said OpenAI would also bring in third-party evaluators with deep internal access. Elon Musk, normally a fierce Altman antagonist, offered an even shorter endorsement: “Dario is right.”

Yet the consensus is fragile. Amodei does not propose stopping development; he argues that tighter safeguards can preserve AI’s benefits while the US maintains an edge over China. Critics of the catastrophe-focused framing counter that the warnings can distract from harms already unfolding—and that rules designed by dominant labs could amount to regulatory capture rather than real accountability.

The test now is whether public declarations become common, independently verifiable limits—or merely another safety pledge in an industry still racing at full speed.

Continue reading https://foxvector.com/stories/01a098a5-7f3b-16f9-7327-3d9b263c7ce6