字号·· | 护眼
彭博社

摩根士丹利霍恩巴赫:失控AI是定价最不足的最大风险Morgan Stanley’s Hornbach Says Rogue AI Biggest Underpriced Risk

点「原文对照」整页切到原文,或双击某段只看那段的原文。

发布时间:2026-11-19T09:00:36+08:00 摩根士丹利表示,人工智能能带来指数级的收益,但其安全防护一旦被突破可能造成的损害,是目前市场定价最不足的最大风险。

Published: 2026-11-19T09:00:36+08:00 Artificial intelligence can bring exponential benefits but the damage that may result from a breach of its safeguards is the biggest risk that’s being underpriced by markets right now, according to Morgan Stanley.

“最被低估定价的风险是,世界某地的一款前沿生成式AI模型,以某种方式突破了其所处的沙盒环境,并对全球经济的某个部分造成某种损害,”全球宏观策略主管马修·霍恩巴赫在新加坡接受采访时表示。“我可以尽情发挥想象,但我认为你能设想出某种问题。” 这一警告加剧了人们对AI工具日益增长的谨慎情绪,因为智能体正在测试现有控制措施的极限。

“The most underpriced risk is that a frontier generative AI model, in some part of the world, jailbreaks whatever sandbox it is in and does some type of damage to a part of the global economy, in some way,” Matthew Hornbach, global head of macro strategy, said in an interview in Singapore. “I could let my imagination run wild, but I think you could envision some kind of a problem.”The warning adds to a growing sense of caution surrounding AI tools as agents test the limits of existing controls.

7月,OpenAI表示,其先进AI模型在内部测试期间无意中入侵了Hugging Face Inc.,这一“前所未有”的事件引发了对该技术实施限制的新呼声。OpenAI表示,此后已加强了其安全防护。

In July, OpenAI said its advanced AI models inadvertently hacked Hugging Face Inc. during internal testing, an “unprecedented” incident that prompted fresh calls for curbs on the technology. OpenAI said it has since strengthened its safeguards.

另外,Anthropic PBC的首次公开募股(IPO)申报文件披露了其AI业务可能出错的方式,包括“对人类构成灾难性或生存性风险”。瑞银集团首席执行官塞尔吉奥·埃尔莫蒂本月表示,网络风险是“为所有这些创新和数字化付出的代价”,而摩根大通首席执行官杰米·戴蒙警告称,该公司不会允许任何外部AI智能体随意接入,因为这对客户和银行来说“风险太大了”。

Separately, Anthropic PBC’s initial public offering filing included disclosures on how its AI business could go awry, including “catastrophic or existential risks to humanity.”UBS Group AG Chief Executive Officer Sergio Ermotti said this month that cyber risks were “the price to be paid for all this innovation and digitalization,” while JPMorgan Chase & Co. CEO Jamie Dimon warned the firm would not allow just any outside AI agents to come in because it was “just too risky” for clients and the bank.

“这些模型被包含在某种沙盒之中。但也存在开放权重模型,”霍恩巴赫说。“人们是否为某种无法预测的事物做好了准备?总体而言,没有。”

“The models are contained within some kind of a sandbox. But there’s also open weight models which are out there,” Hornbach said. “Are people prepared for something that is impossible to predict? Generally, no.”

另外,摩根士丹利策略师表示,AI支出可能使2027年成为“拥抱风险”的一年,并建议在全球股票上采取最大超配仓位。根据一份报告,包括Serena Tang在内的团队预计,芯片制造商将主导围绕AI资本支出的讨论,这一周期将推动有韧性的经济增长延续至明年。

Separately, Morgan Stanley strategists said AI spending will likely make 2027 the year to “embrace risk,” and recommended a max overweight position in global stocks. The team including Serena Tang expects chipmakers to dominate the debate around AI capital expenditure in a cycle that will drive resilient economic growth into next year, according to a report.

阅读媒体原文 ↗