IT之家 10 月 24 日消息,对于 AI 时代的网络安全难题,亿万富翁、企业家兼投资人马克 · 库班提出了一套颇有意思的“灭虫”思路。
据《商业内幕》今天(24 日)报道,库班将大量 AI 智能体发动的网络攻击比作蟑螂入侵:即使房屋刚刚喷过杀虫剂,蟑螂仍能找到钻进去的办法。“我经常问网络安全专家,听到各种黑客攻击、工艺突破和安全难题时,AI 究竟有没有发明什么新招数?答案是没有。AI 并没有创造新的攻击手段,只是可以发动海量攻击,本质上还是暴力破解。”
在库班看来,防范 AI 网络攻击与防止蟑螂入侵房屋有着相似之处。“道理和防蟑螂一样。只要有人制造出足够多的 AI 智能体,总有办法找到平台的入口,再利用找到的漏洞。好消息是,笔者同样可以用大量 AI 给平台‘喷药’。我认为,会有更多人投入力量阻止恶意攻击。我对 AI 一点也不悲观。”
人工智能行业也开始组建自己的“灭虫队”,着手防范安全风险。过去一年,OpenAI 在领英发布的安全岗位招聘数量增长 91%,涵盖平台测试、工艺研究和治理等领域。
IT之家获悉,部分风险已在测试中显现。Anthropic 披露,由于第三方测试环境规格错误,三个 Claude 模型未经授权访问了有关机构的实际运行平台。此后,Anthropic 引入实时分类模型,识别并阻止 AI 模型试图突破测试环境的行为。
库班认为,AI 虽然降低了实施恶意攻击的门槛,但防御者同样可以利用 AI 增强网络安全防护能力。“坏事当然会发生,但网络攻击从来就不是什么新鲜事。如果攻击者能靠大规模暴力破解发动攻击,防御者也能以同样的方式应对。”
本文来源:https://www.ithome.com/1/017/347.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
IT House October 24 news, for the AI era of cybersecurity problems, billionaire entrepreneur and investor Mark Cuban put forward a set of interesting "exterminating" ideas.
According to Business Insider today (24th), Kuban compared cyberattacks by a large number of AI agents to cockroach intrusions: even if the house has just been sprayed with pesticides, cockroaches can still find a way to get in. “I often ask cybersecurity experts whether AI has invented any new tricks when they hear about various hacking attacks, process breakthroughs, and security puzzles. The answer is no. AI has not created new means of attack, but can launch massive attacks, which are inherently violent cracking."
In Cuban's view, preventing AI cyberattacks bears similarities to preventing cockroaches from invading homes. “It's the same as anti-cockroach. As long as someone makes enough AI agents, there is always a way to find the entrance to the platform and then exploit the vulnerabilities found. The good news is that I can also use a lot of AI to ‘spray‘ the platform. I think there will be more people investing in stopping malicious attacks. I'm not at all pessimistic about AI.”
The artificial intelligence industry has also begun to form its own "insect extermination team" to prevent security risks. Over the past year, OpenAI has seen a 91% increase in security job openings posted by LinkedIn, covering areas such as platform testing, process research and governance.
IT Home was informed that some of the risks were already evident in the tests. Anthropic disclosed that three Claude models had unauthorized access to the agency's operational platform due to errors in third-party test environment specifications. Since then, Anthropic has introduced real-time classification models to identify and block the behavior of AI models trying to break through the test environment.
Kuban believes that although AI lowers the threshold for malicious attacks, defenders can also use AI to enhance cybersecurity protection capabilities. “Bad things happen, of course, but cyberattacks are never new. If the attacker can launch an attack with massive brute force, the defenders can respond in the same way.”
This article is from: https://www.ithome.com/1/017/347.htm For copyright issues, please write to: 13951431913#139.com or submit a review on the right.
