中新网北京9月29日电 (陶思阅)今年9月，第22届国家网络安全宣传周网络安全协同防御分论坛活动中， 国内首次针对AI大模型的实网众测检验结果发布。测试累计发现281个安全漏洞，其中大模型特有漏洞 占比超60%，涵盖提示注入、信息泄露等典型风险。 同时，攻击者研究AI的学习偏好，有针对性地"投喂"信息，更有甚者，有组织地进行"数据投毒"，即通 过伪造专家身份、虚构研究报告等方式，向AI输入虚假信息，以达到操纵AI输出结果的目的。 同盾科技安全研究院院长董纪伟认为，随着生成式AI技术的广泛运用，未来的攻防战将迈入"以技术对 抗技术，以AI对抗AI"的阶段。 面对挑战，监管体系率先升级。9月15日，《人工智能安全治理框架》2.0版在国家网络安全宣传周发 布。 AI安全行业也正构建"AI对抗AI"的技术防线。9月，华为发布全新升级的星河AI网络安全解决方案，据 介绍，该方案威胁自动化处置率达99%，未知威胁检出率达95%。 蚂蚁消金通过多模态感知与大小模型协同，系统可精准识别假章假证、合成语音等造假行为。蚂蚁消金 首席信息官刘义介绍，目前，虚假证件识别准确率达98%，语音伪造检测覆盖50余种合成方式，2024年 以 ...

Core Insights - Multiple cybersecurity companies are actively investing in AI technology development, enhancing their product capabilities and operational efficiency [1][2][3] - The integration of AI in cybersecurity is seen as a double-edged sword, presenting both new security risks and opportunities for improved efficiency [1][4] Group 1: Company Developments - Green Alliance Technology plans to launch AI security products, including an AI security integrated machine and a large model security assessment system [1] - North Trust has developed an AI capability platform that integrates large models and development tools, with applications delivered in finance and energy sectors [1][2] - Deepin Technology has incorporated large model technology into its cybersecurity products, including a security GPT and AI firewall, with plans for further investment in AI R&D [2] - Ant Group has released innovative products that combine cybersecurity and AI technology, including a trusted connection framework for smart glasses [2] - Starry Sky Technology's AI model has been applied in security operations and threat detection, significantly enhancing product capabilities [3] - AsiaInfo reported significant growth in AI model applications and deliveries in the first half of the year, focusing on AI model applications, 5G private networks, and intelligent operations [3] Group 2: Industry Trends and Challenges - Gartner's report indicates a shift in focus towards securing AI systems in cybersecurity, with expectations that 60% of large Chinese enterprises will adopt exposure management technology by 2027 [4] - The need for companies to be aware of risks associated with AI model applications, such as prompt injection and model manipulation, is emphasized [4][5] - The importance of supply chain security in AI applications is highlighted, with calls for enhanced version vulnerability management and code security audits [5] - The rapid adoption of AI models is expected to create significant security risks, necessitating a dynamic defense system and cross-departmental collaboration [5][6] Group 3: Recommendations for AI Security - Experts suggest mandatory registration for AI models to identify risks early and ensure comprehensive understanding of their security and usability [6] - Companies are encouraged to conduct compliance assessments and deploy specialized protections, such as AI security barriers, to defend against new types of attacks [6] - Establishing trust through security measures is seen as essential for promoting data flow and maximizing the value of AI applications across various industries [6]

近日，多家网络安全上市公司公布网络安全领域AI技术应用的新进展。绿盟科技、北信源向记者透露 AI创新成果，深信服、亚信安全等企业相继公布应用AI技术的网络安全产品以及市场应用。 行业专家认为，AI技术应用在带来新型安全风险的同时，也为网络安全行业提升效率、提高产品能力 带来机遇。 ● 本报记者 郑萃颖 蚂蚁集团也于近日发布了融合网络安全与AI技术的创新产品及解决方案。其中，智能眼镜可信连接技 术框架gPass已与Rokid、小米、夸克、雷鸟等智能眼镜品牌合作，率先落地"看一下支付"场景。在大模 型安全方面，蚂蚁集团展示了AIGC滥用检测、深度伪造防御在内的多项解决方案。 另外，启明星辰表示，其开发的安星智能体已经应用于安全运营、威胁检测、数据安全等产品或服务 中，显著提升了产品的能力和服务的效率。亚信安全称，上半年，公司AI大模型应用与交付业务实现 显著增长，未来公司将持续聚焦AI大模型应用与交付、5G专网与应用、数智运营三大增长引擎，加快 签单节奏。 AI应用带来安全新挑战 信息技术咨询公司Gartner的报告显示，中国网络安全的重点正逐步转向保障人工智能的安全，实现业 务转型并加强企业、机构的韧性。 "网 ...