2026-11-04 · Fonte：iFUN
The Anthropic chief executive says adding safety mechanisms is no longer enough — the pace of new AI capabilities must fall so testing and safeguards can keep up.
Anthropic chief executive Dario Amodei is calling for a markedly more cautious approach to building powerful AI systems. In his assessment, it is no longer sufficient to bolt additional safety mechanisms onto increasingly capable models. Instead, he argues, the pace at which new capabilities arrive must itself come down, so that evaluation procedures and protective measures have time to keep step with the technology.
The position from the company behind the Claude assistant marks a shift in emphasis: rather than treating safety as an add-on layered over each successive release, the underlying speed of capability advancement should be throttled until verification processes and safeguards can catch up. The discussion directly concerns OpenAI's ChatGPT as well, the other flagship conversational system named in the report.
Amodei is not making the case alone. He has secured backing from OpenAI chief executive Sam Altman and from Elon Musk — an alignment that stands out given how directly the companies and individuals involved compete at the frontier of AI development. When rival leaders converge on the need not merely to add safeguards but to slow the introduction of new capabilities, it signals that the argument over safe AI development has moved beyond questions of feature-level protection.
At its core, the proposal reorders priorities for frontier AI work. Amodei's position holds that testing regimes and risk controls must not be allowed to lag behind model capabilities; where the gap threatens to open, the remedy is not more safety features stacked on top of a fast-moving release cycle, but a deliberate reduction in the speed at which new abilities are introduced. For a field whose commercial momentum rests on rapid iteration and headline-grabbing model upgrades, that is a demanding standard.
Whether that standard is adopted in practice remains the open question left by the report. What is unambiguous is that the leaders behind ChatGPT and Claude now publicly share the view that capability growth has outpaced the means of checking it — and that the tempo of advancement, not only its supervision, should ease.
Tags: Anthropic · OpenAI · Dario Amodei · Sam Altman · Elon Musk · AI Safety
萬安算交所 AIXX · AIXX · Exchange verificado · Ir ao mercado