据CNBC报道，Nvidia首席执行官黄仁勋表示，人工智能公司必须对确保其模型安全负责，而不应等待政府出台新法规。
Nvidia Chief Executive Jensen Huang said artificial intelligence companies must take responsibility for making their models safe rather than wait for governments to introduce new regulations, CNBC reported.
在Nvidia于柏林举行的GTC大会上发言时，黄仁勋说，公司应立即应对已知风险，因为立法需要时间制定和实施。
Speaking at Nvidia’s GTC conference in Berlin, Huang said companies should address known risks immediately because legislation takes time to develop and implement.
“如果今天存在担忧，你们有责任承担问责并立即采取行动，而不是等待法律法规出台，那需要时间，”黄仁勋说。
“If there are concerns that exist today, it is your responsibility to take accountability and do something about it now, instead of waiting for laws and regulations to happen, which takes time,” Huang said.
黄仁勋认为，公司也有商业动力构建安全产品，称安全是其成功的关键。
Huang argued that companies also had a commercial incentive to build safe products, saying safety was essential to their success.
“不存在不安全的好产品，”他说，并补充道，企业有理由负责任地开发和部署AI。
“There’s no such thing as a good product that’s not safe,” he said, adding that businesses had reasons to develop and deploy AI responsibly.
Nvidia的OpenShell如何旨在遏制AI代理 黄仁勋指出OpenShell——Nvidia用于遏制和监控AI代理的技术——是随着AI系统更具备自主行动能力而管理风险的一种方式。
How Nvidia’s OpenShell aims to contain AI agents Huang pointed to OpenShell, Nvidia’s technology for containing and monitoring AI agents, as one way to manage risks as AI systems become more capable of acting autonomously.
该技术旨在监控AI代理的行为并限制其访问权限，帮助防止它们采取未经授权的行动。黄仁勋将这种方法比作将婴儿置于围栏中，孩子可以在规定边界内活动，同时仍处于监督之下。
The technology is designed to monitor what AI agents do and restrict their access, helping prevent them from taking unauthorised actions. Huang compared the approach to keeping a baby in a playpen, where the child can move around within defined boundaries while remaining under supervision.
他将OpenShell描述为更广泛安全体系的第一层，该体系将需要额外的防护措施。
He described OpenShell as the first layer of a broader safety system that would require additional protections.
“OpenShell是第一层，然后是Sentry，第二层，接着你们将拥有多层防御，”黄仁勋说。
“OpenShell is the first layer, and then there’s Sentry, the second layer, and then you’re going to have multiple layers of defence,” Huang said.
黄仁勋对该技术解决涉及非预期访问的某些事件的能力表示信心，同时承认行业无法预见所有未来风险。“我们不知道未来有哪些未知的未知，”他说。
Huang expressed confidence in the technology’s ability to address certain incidents involving unintended access, while acknowledging that the industry could not anticipate every future risk. “We don’t know what we don’t know about the future,” he said.
黄仁勋表示，全球有超过100家公司参与了OpenShell计划。这些公司包括Anthropic、OpenAI、SpaceX、Meta Platforms、Google、Microsoft、IBM和Red Hat，以及网络安全公司CrowdStrike和Palo Alto Networks。
More than 100 companies involved in OpenShell initiative Huang said more than 100 companies worldwide were part of the OpenShell initiative. They include Anthropic, OpenAI, SpaceX, Meta Platforms, Google, Microsoft, IBM and Red Hat, as well as cybersecurity companies CrowdStrike and Palo Alto Networks.
黄仁勋说，该计划汇集了操作系统提供商、网络安全公司和先进AI模型开发商，共同致力于安全措施。他补充说，Anthropic是OpenShell的早期采用者，并已与英伟达在该技术上展开合作。
The initiative brings together operating-system providers, cybersecurity firms and developers of advanced AI models to work on safety measures, Huang said. He added that Anthropic was an early adopter of OpenShell and had collaborated with Nvidia on the technology.
黄仁勋表示，更广泛的目标是确保AI发展在负责任地推进的同时创造一个乐观的未来，支持现有产业并催生新业务。
Huang said the broader goal was to ensure that AI development created an optimistic future while being pursued responsibly, supporting existing industries and enabling new businesses.
他说：“我们作为一个行业最重要的任务，是创造一个乐观的未来，但要负责任地去做。”
“The most important thing that we need to do as an industry is to create an optimistic future, but do it responsibly,” he said.
他的这番言论发表之际，AI公司因如何管理日益强大的模型带来的风险而面临越来越严格的审视，特别是那些能在有限人工监督下执行任务的系统。
His remarks come as AI companies face growing scrutiny over how they manage risks from increasingly capable models, particularly systems that can carry out tasks with limited human supervision.