军事·安全
专家警告:目前,人工智能领域的几乎所有人都有可能陷入诱骗陷阱。
中文译文
尽管英伟达因黄仁勋涉嫌设下诱捕陷阱的丑闻而引发的后果仍在持续,硅谷却尚未面对一个新现实:更多人正面临风险。
“没人会向创业者介绍反情报知识,”前中央情报局高级官员、现任软件公司Lumbra的创始人亚伦·布朗对TechCrunch表示。
这是因为从历史上看,诱捕陷阱通常与政治家、军队以及国家情报机构有关——这些机构拥有安全许可,其员工也受过训练,能够识别间谍和其他攻击者的企图。
但布朗指出,如今许多最宝贵的机密都“存在于那些研发芯片、模型和基础设施的公司内部”,他补充说,间谍“想要的是CEO能够接触到的东西:发展路线图、出口决策、客户名单,以及谁会获得这项技术。CEO是通往这一切的最短路径。”黄某被指控因与艾丽斯·黄发生关系而落入此类美人计。艾丽斯·黄曾是新加坡一家云计算公司的高管,她被指控为中国间谍,并试图将价值数十亿美元的受限制Nvidia芯片走私到中国。
关于黄某的指控曝光后,网上充斥着关于下一位科技行业CEO会是谁的玩笑,但面临风险的并非只有CEO和创始人。《华尔街日报》报道称,OpenAI开始警告员工,有人会试图接近他们以获取公司机密。从董事会成员到执行助理,任何能够接触敏感信息的人都有可能成为美人计的目标。
布朗表示,通常情况下,实施美人计的人并非恋人,友谊实际上才是最古老的美人计手段。这些美人计以投资者、客户、招聘人员或咨询顾问的身份接近目标对象。他们的请求在表面上看起来很合理,无需让目标对象产生好感。“他们希望你能放心地与之交谈,”布朗说道,“因为没人会留意,所以这种手段能持续更长时间。”
有趣的是其中的讽刺之处:许多科技行业创始人致力于将人类生活的诸多方面实现自动化和数字化,却依然容易受到最古老的骗术的侵害——那种利用人们希望被他人理解的欲望来行骗的手段。
“既然可以接近那些能够打开大门、影响决策或提出没人愿意质疑的要求的人,何必还要费力突破防火墙呢?”KnowBe4公司的犯罪学家兼首席反欺诈策略师佩里·卡彭特对TechCrunch说道。
一旦目标对象意识到自己已被欺骗,威胁程度可能会升级:攻击者此时便有了敲诈的手段。另一个讽刺之处在于:人工智能让诱骗行为变得更简单、成本更低。过去,策划一场骗局需要数月甚至数年的研究,包括塑造伪装身份、了解目标对象以及实施计划。如今,利用人工智能就能轻松研究目标对象,创建虚假的社交资料、伪造背景,甚至进行虚假的语音通话。
“人工智能可以帮助构建一个完整的虚构背景故事,”卡彭特说。“这项技术有助于铺垫场景,但最终的欺骗行为还是通过人际关系来实现的。”
布朗表示,人工智能行业“发展成国家安全产业的速度,远远快于其行业文化的成熟速度”。
“被诱骗的部分原因就在于当事人并不知情,”他补充道。“这并不让人感觉像是被操纵,反而会让人觉得是真实的,甚至是幸运的。”卡彭特表示,一个好的第一步是留意那些不寻常的迹象。例如,他说要注意“那些发生频率稍高的巧合性相遇”。此外还有一些明显的信号,比如对工作及同事持续表现出好奇心,或者要求对会面保密,还不断要求介绍人或获取信息。
更严重的警示信号包括,当某人的要求越来越过分,超出你通常会感到不适的界限时。“只需凭直觉判断,”布朗说,“如果你明天失业了,这个人还会留在你身边吗?”
尝试理解情绪何时可能影响决策也很有帮助。卡彭特建议在个人生活与工作生活之间划清明确的界限。他在公司内部还指出,应“消除高管特权观念”,即那种认为规则不适用于公司里掌权人士的思维方式。布朗补充说,企业文化也起着重要作用:员工是否愿意提出担忧、质疑不合理的请求,并上报潜在的威胁?卡彭特表示:“最危险的安全文化就是那种没人敢质疑首席执行官的文化。”
布朗说:“创始人需要承担风险,也需要探索商业机会。偏执毫无帮助,而意识到自己可能是目标则很有意义。”
原文
Even as the fall-out for Nvidia over Jensen Huang’s alleged honey-trap scandal is still unfolding, Silicon Valley has yet to reckon with a new reality: many more people among them are at risk.
“Nobody briefs founders on counterintelligence,” Aaron Brown, a former senior CIA officer and current founder of the software company Lumbra, told TechCrunch.
That’s because historically honey traps have been associated with politicians, the military, and national intelligence — the people “behind clearances and a workforce trained to expect an approach,” from spies and other attackers.
But now many of the most valuable secrets “sit inside companies building chips, models, and infrastructure,” Brown says, adding that a spy “wants what the CEO can touch. The roadmap, the export decisions, the customer list, who gets the technology. The CEO is the shortest path to all of it.”
Huang has been accused of falling for just such a honey trap by having an affair with Alice Huang. She’s a former executive of a Singapore neocloud who’s been accused of being a Chinese spy and of smuggling billions of dollars’ worth of restricted Nvidia chips into China.
After the Huang allegations broke, jokes flooded the internet about which tech CEO will be next, but it’s not just CEOs and founders who are at risk. The WSJ reported that OpenAI started warning its employees about outsiders trying to cozy up to them in search of company secrets. Anyone with access to sensitive information — from board members to executive assistants — can be the target of a honey trap.
And typically, the attackers aren’t lovers. Friendship is actually the oldest honey-trap scheme around, Brown said.
Honey traps gain access to their targets as investors, customers, recruiters, and consultancies. Their requests look legitimate on paper and don’t need their targets to fall in love. “They need you to feel comfortable talking,” Brown said. “They last longer because nobody’s looking.”
What’s interesting, too, is the irony. So many tech founders are moving to automate and digitize so much of human life and yet remain vulnerable to the oldest bug around, a person who preys on the human need to be understood by others.
“Why fight through a firewall when you can get close to the person who can open doors, influence decisions, or make requests nobody wants to question?” Perry Carpenter, a criminologist and chief deception strategist at KnowBe4, told TechCrunch.
Once targets realize they have been compromised, the threat can escalate: attackers now have a means of blackmail.
Another irony: AI has made honey-trapping much easier and cheaper. Planning a scheme used to take months or years of research, crafting a persona, knowing the target, and executing a plan. These days, it’s easy to use AI to research a target, generate fake social profiles, fake backgrounds, even fake voice calls.
“AI can help sustain an entire fictional backstory,” Carpenter said. “The technology helps set the stage, but the deception ultimately plays out through human relationships.”
The AI industry “became a national security industry faster than its culture caught up,” Brown said.
“Part of being honey-trapped is not knowing,” he added. “It doesn’t feel like manipulation. It feels genuine, even lucky.”
Carpenter said a good first step is looking for unusual patterns. Note, for instance, “coincidental encounters that happen a little too often,” he said. There are also obvious signs, like persistent curiosity about work and colleagues, or requesting secrecy about meeting up, and pressing for introductions or information.
Bigger red flags include if someone’s requests are getting bigger, pushing the boundaries of what you usually would be uncomfortable doing. “A simple gut check,” Brown said. “Would this person still be around if you lost your job tomorrow?”
It also helps to try and understand when emotions might be influencing a decision. Carpenter suggests putting up a clear boundary between one’s personal and professional life. Within companies, it also helps to “eliminate executive exceptionalism,” he said, the idea that the rules don’t apply to people in power at the company.
Brown added that company culture plays another role, too: Are employees comfortable with raising concerns, challenging unusual requests, and escalating possible threats? “One of the most dangerous security cultures is one where nobody feels comfortable questioning the CEO,” Carpenter said.
“Founders need to take risks, and they need to explore business opportunities,” Brown said. “Paranoia doesn’t help. Recognizing you’re a target does.”
