内容来源：https://aibusiness.com/agentic-ai/the-growing-need-for-cybersecurity-in-agentic-ai

内容总结：

谷歌云赞助播客节目探讨企业生成式AI应用安全新挑战

在近期播客节目《Targeting AI》中，AI数据与访问管理初创企业Barndoor.ai创始人兼CEO奥伦·米歇尔斯指出，企业在部署生成式AI系统时需建立区别于传统IT安全的全新防护范式。该节目于纽约AI峰会现场录制，由主持人埃斯特·希图主持。

米歇尔斯强调，随着AI智能体在企业内部普及，员工正借助其完成工作任务，但AI行为结果存在不确定性。传统网络安全侧重于防范外部入侵，而AI时代的安全防护需从"以人为中心"转向"以智能体为核心"，在验证使用者身份的同时，更需对AI代理行为本身实施安全管控。

节目同时建议，企业引入生成式AI应优先选择能优化信息交互体验的应用场景，避免因系统部署不当引发安全风险。此次讨论为企业构建AI时代安全体系提供了重要思路。

中文翻译：

由谷歌云赞助
选择您的首个生成式AI应用场景
要迈出生成式AI的第一步，首先应关注那些能够优化人类信息交互体验的领域。企业在实施AI系统时常常面临挑战，决策失误往往使其暴露于安全威胁之下。

正如任何IT系统一样，商业环境中的生成式AI与智能体AI部署必须确保安全性。在AI Business推出的《Targeting AI》播客最新一期中，AI数据与访问管理初创公司Barndoor.ai创始人兼首席执行官奥伦·米歇尔斯，深入探讨了企业级AI系统快速扩张带来的网络安全影响。

本期由埃斯特·希图主持的播客录制于纽约AI峰会现场，米歇尔斯原定在此参与关于企业级AI安全的专题讨论。他在节目中指出，企业需要针对智能体系统采取不同于传统IT网络安全策略的全新防护思路——旧有模式侧重于防止外部入侵者渗透系统并造成破坏。

"如今，AI智能体正被组织内部人员用于推进和完成既定任务，"米歇尔斯解释道，"但这些AI智能体的行为未必完全符合预期目标。因此我们必须确保：安全机制应建立在智能体行为本体之上，不仅验证使用者身份，更要全面考量智能体自身的运行特性。"

您可能还喜欢

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
Businesses frequently struggle with implementing AI systems, often making mistakes that leave them vulnerable to security threats.
As with any IT system, generative and agentic AI implementations in business settings need to be secure.
On the latest episode of the Targeting AI podcast from AI Business, Oren Michels, founder and CEO of AI data and access management startup Barndoor.ai, discusses the cybersecurity implications of the rapidly proliferating array of new enterprise AI systems.
The podcast episode, hosted by Esther Shittu, was recorded live at the AI Summit in New York City, where Michels was scheduled to appear on a panel to discuss enterprise-grade AI security.
On the podcast, Michels said enterprises need to take a security approach to agentic systems that differs from traditional IT cybersecurity strategies. The old paradigm focused on preventing outsiders from infiltrating the system and causing harm.
Now, with AI agents, people from inside the organization are using agents to pursue and complete tasks they're supposed to do.
"But the AI [agents] might or might not do the things that are expected for those outcomes to happen," Michels said. "So, we want to make sure that security is provided where the agency is based, not just on the identity of the human, but also takes into account aspects of the agent itself."
You May Also Like