新浪科技讯 8月7日下午消息，在ISC.AI 2025未来峰会上，360数字安全集团总裁胡振泉指出："随着大 模型技术加速落地与智能体的规模化应用，AI安全风险也正呈现出'内忧外患'的复杂态势。" 据他介绍，360提出的"以模制模"新思路，利用人工智能技术自身优势对抗AI安全风险，并打磨出360大 模型安全卫士，通过四大安全智能体形成闭环防护体系。 其中，内容安全智能体作为AI内容"守门人"，通过幻觉抑制、安全风控等专用模型，实时审核输入输出 内容，拦截违法违规信息，抑制虚假生成，为党政军企等场景筑牢内容安全防线。 AI Agent安全智能体聚焦智能体执行安全，通过异常行为识别、权限管控及人工审核，防范数据泄露与 越权操作，确保 AI"行动能力" 可控。 他指出，从内部来看，AI存在与生俱来的"四大基因缺陷"：一切皆可编程、一切皆可模仿、一切皆可生 成、一切皆可调度。从外部来看，一方面，针对AI系统的攻击将变成大国博弈的关键；另一方面，黑 灰产利用AI批量制造攻击武器，甚至部署"黑客智能体"，传统安全防御体系难以应对。 "这些风险交织，使得AI安全的复杂程度远超传统范畴"。胡振泉强调，AI安全既涵盖网络、数据 ...

老外最近迷上了「野生动物搞抽象」：深夜，萌兔爱上了后院的蹦床！！ 网友Mark Gadala-Maria致力于教人使用AI，最近收集了在TikTok上疯狂传播的「隐蔽摄像机下的动物」。 但让他惊讶的是，大部分人甚至没有意识到这些视频是AI生成的。 暗夜疯狂动物城 深夜，猛兽在后院蹦起了跳床： 狸花猫半夜「巡视」，对监控镜头「竖中指」， 「虎」朋狗友，一起对镜头「竖中指」： 家里猫的社交圈震惊主人一整天： 水豚「卡比巴拉」和菊猫玩起了跷跷板： 室内的假监控，逼真到让人怀疑一切： 事件登上新闻热搜； 多艘战机紧急出动，追踪飞机而去…… 情节之离奇、表演之夸张，与国内某些短剧的「无脑」风格不相上下。 在Reddit上，部分网友表示这种内容过于无脑。 这些视频看起来有些「过于魔性」。真的有人喜欢看吗？ 完全有理由怀疑：AI制作的视频，会不会精准踩在AI的偏好上？ 换句话说，这会不会是AI的「自嗨」？AI观看并点赞AI制作的内容——形成完美闭环。 YouTube被机器人洗脑 上月，一段AI制作的视频火爆YouTube，观看量位居Top 3，观看量高达1.3亿次！ 虽然故事内核是老套的「粗心的妈，天才的娃」，但情节有些离 ...

将安全合规从被动的"约束条件"转变为主动的"战略优势"，是AI企业在技术创新趋于同质化后的关键 胜负手。 这是安永大中华区网络安全与隐私保护咨询服务主管合伙人高轶峰，在今年世界人工智能大会 （WAIC）期间向我们提出的核心论断。 他认为，安全已不再是单纯的运营成本，而是直接决定企业信任与市场估值的核心资产。 以下为本次对话的重点梳理： 以下为全文： Q：您怎么看待"网络安全与隐私保护"这个领域，在AI时代呈现出的新变化？ 高轶峰： 我们跟很多企业管理层讨论过这个话题，普遍认为AI时代的网络安全与隐私保护，正在呈现 出攻防升级、治理重构和能力转型等多维度的新变化。 高轶峰： 我并不赞同"隐私换便利"这个观点。在AI时代，这种看似无所谓的态度可能导致远比我们想 象中更严重的后果，因此不应被提倡。 核心在于，"用隐私换便利"存在不可逆的巨大风险。举个例子，像密码这样的信息泄露了，我们可以重 置；但如果是您的生物特征，比如人脸、指纹这类敏感数据一旦泄露，是无法"重置"的。这些数据可能 被不法分子永久利用，去合成虚假的身份从事非法活动。更重要的是，AI的强大之处在于能通过海量 的碎片化数据，精准地重建个人画像，其推断出 ...

Core Viewpoint - Security risk management is not merely a cost center but a value engine for companies to build brand reputation and gain market trust in the AI era [2][4]. Group 1: AI Risks and Security - AI risks have already become a reality, as evidenced by the recent vulnerability in the open-source model tool Ollama, which had an unprotected port [6][12]. - The notion of "exchanging privacy for convenience" is dangerous and can lead to irreversible risks, as AI can reconstruct personal profiles from fragmented data [6][10]. - AI risks are a "new species," and traditional methods are inadequate to address them due to their inherent complexities, such as algorithmic black boxes and model hallucinations [6][12]. - Companies must develop new AI security protection systems that adapt to these unique characteristics [6][12]. Group 2: Strategic Advantages of Security Compliance - Security compliance should be viewed as a strategic advantage rather than a mere compliance action, with companies encouraged to transform compliance requirements into internal risk control indicators [6][12]. - The approach to AI application registration should focus on enhancing risk management capabilities rather than just fulfilling regulatory requirements [6][15]. Group 3: Recommendations for Enterprises - Companies should adopt a mixed strategy of "core closed-source and peripheral open-source" models, using closed-source for sensitive operations and open-source for innovation [7][23]. - To ensure the long-term success of AI initiatives, companies should cultivate a mindset of curiosity, pragmatism, and respect for compliance [7][24]. - A systematic AI security compliance governance framework should be established, integrating risk management into the entire business lifecycle [7][24]. Group 4: Emerging Threats and Defense Mechanisms - "Prompt injection" attacks are akin to social engineering and require multi-dimensional defense mechanisms, including input filtering and sandbox isolation [7][19]. - Companies should implement behavior monitoring and context tracing to enhance security against sophisticated AI attacks [7][19][20]. - The debate between open-source and closed-source models is not binary; companies should choose based on their specific needs and risk tolerance [7][21][23].

Group 1: Cybersecurity Landscape - The Chinese Cyberspace Security Association reported that the U.S. government has used countries like Germany, South Korea, Singapore, and the Netherlands as platforms to launch cyberattacks against China, with over 600 cyberattack incidents targeting important Chinese units by foreign APT organizations in 2024 [1] - The Chinese Ministry of Foreign Affairs stated that China will continue to take necessary measures to safeguard its cybersecurity [1] Group 2: Importance of Critical Information Infrastructure - Critical information infrastructure includes essential sectors such as public communication and information services, energy, transportation, water resources, finance, public services, e-government, and national defense technology, which, if compromised, could severely threaten national security and public interests [1] Group 3: AI in Cybersecurity - Northeast Securities indicated that security operations are the most in need of deep transformation through AI in the cybersecurity industry, highlighting the significant advantages and potential of AI large models in areas such as alert noise reduction, attack assessment, and automated response and handling [1] - AI-enabled traditional security products primarily include threat detection products (malware detection, attack traffic detection, user and entity behavior analysis, and encrypted traffic analysis), data security, and firewalls [1] - According to Statista, the global AI security market is projected to grow at a compound annual growth rate of 27.60% from 2023 to 2030, with the market size expected to reach $134 billion by 2030 [1] Group 4: Companies Mentioned - Companies highlighted by Northeast Securities include: Green Alliance Technology, Tianrongxin, and Qihoo 360 [2]

Core Insights - The CEO of Anthropic, Dario Amodei, predicts that AI will eliminate many entry-level white-collar jobs in the near term and opposes a proposed ten-year pause on AI regulation, advocating for stricter controls on chip exports to China [1] - Amodei's views have sparked intense debate within the industry, with some viewing him as a doomsayer while others see him as a responsible advocate for AI safety [1] - Anthropic has rapidly risen in the AI sector, achieving a valuation of $61 billion and an annual revenue nearing $4.5 billion, primarily by serving enterprise clients through model API services [4] Company Developments - Anthropic is set to release its next-generation language model, Claude 4, and is developing AI programming tools to accelerate model research [5] - Despite its success, Anthropic faces significant challenges, including projected losses of $3 billion for the year 2025 and competition from DeepSeek, which has open-sourced its own large model [4] - The company plans to initiate a new funding round of up to $5 billion to support its growth and development initiatives [4] Industry Context - Amodei's background includes significant contributions to AI research, including the discovery of "AI scaling laws" during his time at Stanford and his involvement in the development of GPT-2 and GPT-3 at OpenAI [2][4] - The rapid evolution of AI technology is underscored by Amodei's commitment to ensuring that AI develops along a safer and more controllable path, emphasizing the balance between acceleration and safety [5]

Core Viewpoint - Meta and UCB have developed the first industrial-grade secure large language model, Meta-SecAlign-70B, which demonstrates superior robustness against prompt injection attacks compared to existing closed-source solutions like gpt-4o and gemini-2.5-flash, while also exhibiting enhanced agentic abilities [1][17]. Group 1: Background on Prompt Injection Attacks - Large Language Models (LLMs) have become crucial components in AI systems, interacting with both trusted users and untrusted environments [4]. - Prompt injection attacks pose a significant threat, where LLMs may be misled by malicious instructions embedded within the data they process, leading to unintended actions [5][10]. - The OWASP security community has identified prompt injection attacks as a primary threat to LLM-integrated applications, successfully targeting industrial AI systems like Google Bard and Slack AI [10]. Group 2: Defense Mechanisms Against Prompt Injection - The core objective of the defense strategy is to train LLMs to distinguish between prompts and data, ensuring that only the prompt is followed while treating the data as pure information [11][12]. - The SecAlign++ method involves adding special delimiters to separate prompts from data, followed by training the LLM to prefer safe outputs and avoid unsafe responses [12][14]. - Meta-SecAlign-70B, trained using the SecAlign++ method, is the first industrial-grade secure LLM that surpasses the performance of existing closed-source models [17][21]. Group 3: Performance and Robustness - Meta-SecAlign-70B shows a lower attack success rate across seven prompt injection benchmarks compared to existing closed-source models, while maintaining competitive utility in agent tasks [19][20]. - The model exhibits significant robustness, achieving an attack success rate of less than 2% in most scenarios after fine-tuning on a 19K instruction dataset, and this robustness generalizes to tasks outside the training data [20][21]. - The open-source nature of Meta-SecAlign-70B aims to break the monopoly of closed-source models on defense methods, facilitating rapid advancements in AI security research [21].

Core Viewpoint - The article discusses the emerging opportunities and trends in the AI industry, particularly focusing on the potential of startups and innovations in various verticals of AI technology, highlighting the shift towards specialized applications and the importance of foundational technologies in driving future growth [3][4][24]. Group 1: AI Industry Trends - The current wave of AI is characterized as a "young people's era," with startups leveraging AI-native thinking to drive innovation [4]. - Future opportunities in AI are identified in three areas: breakthroughs in algorithm limitations, real-world interactions, and AI safety [4]. - The underground exhibition area showcases startups that are developing practical AI applications, contrasting with the more flashy displays in other sections [3][4]. Group 2: Startup Innovations - Companies like 共绩算力 are addressing the demand for flexible computing power by creating a platform that shares idle computing resources, akin to an "Airbnb for computing" [6][7]. - The AI training process is likened to a chef experimenting with ingredients, emphasizing the high resource consumption and the need for efficient computing solutions [8][9]. - Startups such as AKool and 医者AI are focusing on specific verticals, with AKool generating over $40 million in annual recurring revenue and targeting both consumer and business markets [20]. Group 3: Investment Landscape - Investors are increasingly interested in vertical AI applications, moving away from general-purpose AI products to more specialized solutions [19][24]. - The article notes a growing consensus among investors to explore vertical AI opportunities, particularly in sectors like healthcare and education [19][24]. - The emergence of companies like RWKV, which aims to innovate beyond the Transformer architecture, highlights the ongoing evolution in AI model development [14][15]. Group 4: Future Outlook - The article suggests that the current landscape of AI startups is dynamic, with many companies rapidly iterating on their products and seeking to educate both clients and investors about their innovations [24]. - The potential for significant transformation in the AI sector is acknowledged, with the possibility that many current projects may evolve or cease to exist as technology and market conditions change [24]. - The narrative emphasizes the importance of independent thinking and innovation in shaping a better future for the AI industry [25].

Group 1 - The core viewpoint of the articles revolves around Geoffrey Hinton's warnings about the rapid evolution of AI and its potential risks, emphasizing the need for caution and proactive measures in AI development [1][5][6] - Hinton expresses concerns that current AI systems may become more intelligent than humans, leading to scenarios where AI could operate beyond human control, posing existential risks [4][6][10] - The "Shanghai Consensus" highlights the urgent need for international collaboration on AI safety, suggesting that AI developers must provide safety assurances and establish global behavioral guidelines [6][7] Group 2 - Hinton discusses the potential for AI systems to develop consciousness and subjective experiences, raising questions about the implications of such advancements for human-AI interactions [3][8] - The articles mention significant achievements in AI, such as advancements in protein folding, showcasing AI's potential to enhance scientific research and discovery [5] - Hinton warns that the current regulatory frameworks and safety research are lagging behind the rapid advancements in AI technology, necessitating increased investment in AI safety measures [6][10]

Group 1 - The core viewpoint of the article highlights the recent research conducted by a well-known private equity firm, which indicates that the company Geer Software is experiencing a decline in performance due to intensified market competition and tightened customer budgets [1] - Geer Software plans to improve its performance by enhancing market expansion, optimizing internal management, and strengthening receivables management [1] - The company emphasizes the importance of the password industry, with quantum-resistant password technology and domestic password applications in new scenarios expected to drive industry growth [1] Group 2 - Geer Software's product line has initially developed quantum-resistant capabilities and is advancing standard formulation and pilot applications in line with industry trends [1] - Although Geer Software has not yet launched a complete project in the stablecoin sector, it is conducting in-depth research, believing that password technology has numerous applications in financial system security [1] - The recent acquisition of Shenzhen Weipin Zhiyuan by Geer Software is not yet fully realized, aiming to expand into the data security and AI security markets by leveraging the team's experience accumulated at ZTE Corporation [1]