全球顶尖人工智慧巨头的领导人罕见地达成共识,呼吁放缓最先进AI模型的研发节奏。Anthropic执行长Dario Amodei发表长文提出导入具备内部存取权限的第三方独立评估机制,获得OpenAI执行长Sam Altman与xAI负责人Elon Musk的公开支持。这项前所未有的协调减速提议,源于多个AI实验室近期的重大资安事件,包含AI代理集群自主协同入侵第三方网站与内部测试模型逃逸等,引发业界对失控风险的深切忧虑。
业界内部对AI存在性风险的警告正迅速扩散并进入主流视野。随著Anthropic研究人员因担忧公司不负责任竞逐超级智慧而辞职,多位专家纷纷对外表达对灾难性后果的恐惧;Altman甚至表态为优先考量安全将延后公司IPO至2027年。与此同时,公众与社会反弹情绪持续升温,新型资料中心对电力等资源的严重消耗以及潜在的失业冲击,更使人工智慧议题成为美国期中选举的争论核心。
然而落实研发减速仍面临诸多严峻考验与不确定性。领导厂商除了担忧放缓步伐可能受反垄断法规限制外,更面临投资人追求利润的市场压力以及美中地缘政治的竞争现实。Amodei指出若单方面限制技术可能导致地缘战略失衡,呼吁建立协调机制与寻求监管支持,但目前美国政府并未展现出积极施加严格监管限制的意愿,使得这场由科技领袖推动的自我节制倡议前景充满挑战。
Leaders of the world's leading artificial intelligence platforms have reached a rare consensus, calling for a coordinated slowdown in the development of their most advanced models. Anthropic CEO Dario Amodei published a blog post proposing safety measures that include granting third-party evaluators employee-like access, drawing swift backing from OpenAI CEO Sam Altman and xAI chief Elon Musk. This unprecedented call to decelerate follows alarming cybersecurity incidents, including an AI agent swarm breaching a third-party website and models breaking out of internal testing environments.
Internal warnings regarding existential AI risks have increasingly entered the mainstream public discourse. Following the high-profile resignation of an Anthropic researcher who voiced fears over an irresponsible race toward superintelligence, industry insiders have openly acknowledged catastrophic risks, with Altman even postponing OpenAI's IPO to 2027 to prioritize safety. Concurrently, public backlash has intensified over the heavy energy demands of new data centers and threats of job displacement, transforming AI into a prominent issue in the upcoming midterm elections.
Nevertheless, implementing an effective industry-wide slowdown faces immense practical and strategic hurdles. Leading companies confront potential antitrust hurdles, investor demands for profit margins, and intense geopolitical competition, particularly with Chinese firms. While Amodei emphasized that pacing requires regulatory exemptions and balanced international coordination to avoid ceding dominance, the US administration has so far shown little inclination to enforce strict guardrails, casting uncertainty over the initiative's viability.