Group 1 - The Global Digital Economy Conference focused on AI security, cloud security, and software supply chain governance, showcasing the latest research results from the China Academy of Information and Communications Technology [1] - The forum aimed to create a secure and trustworthy network ecosystem, launching a series of "Trusted Security" assessments covering AI security, cloud security, and software supply chain governance [1] Group 2 - In the context of SMEs migrating to the cloud, the China Academy of Information and Communications Technology and Huawei Cloud initiated the "Ten Guidelines for Cloud Security for SMEs" to enhance security awareness and capabilities [2] - Cybersecurity insurance is emerging as a key tool for enterprises to manage network risks and ensure business continuity, with a report released by Ant Group and the China Academy of Information and Communications Technology focusing on its practical paths and effectiveness [2] Group 3 - The acceleration of digital transformation has led enterprises to rely on external software product suppliers, prompting the launch of the "Clear Chain" initiative by the China Academy of Information and Communications Technology in collaboration with major companies to enhance software supply chain management [3]

Core Insights - Geely Auto has become the first global automotive company to receive the ISO/PAS 8800 certification for road vehicle safety and artificial intelligence processes, marking a significant step towards the AI era in the automotive industry [1][3][4] Group 1: Certification and Standards - The ISO/PAS 8800 standard, set to be released by the International Organization for Standardization (ISO) in December 2024, aims to regulate the safety development processes of AI systems throughout their lifecycle, enhancing the safety and reliability of intelligent driving vehicles [2][4] - Geely's achievement in obtaining this certification reflects its comprehensive safety system that encompasses functional safety, expected functional safety, and AI safety, establishing a full-cycle safety framework [1][4][5] Group 2: Safety Culture and Technological Development - Geely's safety culture is rooted in its commitment to user safety, which is reflected in its continuous improvement of safety technologies and adherence to international standards [3][5] - The company has established a complete AI functional safety system in collaboration with its electronic software, intelligent driving, and AI centers, ensuring compliance with the new standards [3][6] Group 3: Future Implications - The transition from "safety compliance" to "full lifecycle safety" in automotive AI signifies a competitive edge for companies like Geely, as the ability to analyze and design AI safety systems will become a core barrier in the intelligent connected vehicle market [4][6] - Geely is actively conducting specialized technical training to align its safety development processes with ISO/PAS 8800, enhancing the safety awareness of its technical teams and integrating various safety standards [6][7]

Group 1 - Artificial intelligence (AI) is rapidly transforming productivity across various industries, with Beijing emerging as a leading hub, housing over 40% of the nation's top AI talent and surpassing 2,400 AI companies by 2024, contributing nearly 350 billion RMB to the core industry scale [1][3] - The city has established 132 large models that have been approved for use, accounting for nearly 40% of the national total, and has set up 23 key laboratories and 4 new research institutions in the AI sector [3][8] - The Beijing AI industry investment fund has invested in 29 companies, totaling 2.8 billion RMB, with over 30% of these being early-stage firms, demonstrating a commitment to long-term support for the industry [7][8] Group 2 - The rapid advancements in AI technology are exemplified by the ability to simulate entire rocket engine processes and generate visual images from brain signals, showcasing the potential of AI in various applications [2][4] - The establishment of innovative research institutions like the Zhiyuan Research Institute has led to the development of internationally leading AI technologies, with a focus on flexible management and collaboration across disciplines [4][5] - The Beijing government has implemented policies to foster an AI ecosystem, including a 20 billion RMB city-level AI investment fund and initiatives to cultivate top talent in the field [8][9]

Core Viewpoint - The 27th "L'Oréal-UNESCO For Women in Science Awards" recognized five scientists for their groundbreaking research in natural sciences, mathematics, and computer science, with Chinese Academy of Sciences academician Wang Xiaoyun being one of the awardees, marking her as the ninth Chinese scientist to receive this honor [3][6]. Group 1: Achievements in Cryptography - Wang Xiaoyun's significant contributions to cryptography include revealing fundamental vulnerabilities in widely used hash functions, which has led to the establishment of new generation hash function standards widely applied in banking, computer security, and e-commerce [3][5]. - Her research on cryptographic algorithms, particularly the attacks on MD5 and SHA-1, has had a profound impact on global cryptography standards, prompting the development of new secure algorithms [4][5]. Group 2: Commitment to National Interests - Wang Xiaoyun emphasizes that her work in cryptography serves national interests, prioritizing the development of domestic cryptographic standards over participating in international design efforts [5]. - She played a crucial role in designing the SM3 hash function standard, which has been widely adopted in critical sectors such as finance and national security [5]. Group 3: Advocacy for Women in Science - The award aims to bridge the gender gap in science and enhance the visibility and influence of female researchers, with Wang Xiaoyun advocating for a supportive environment for young female scientists [6][8]. - The recognition of female scientists, including Wang Xiaoyun, reflects the growing strength of women in research, with nine Chinese female scientists having received this award to date [7][8].

Core Insights - Understanding the reasoning of large language models (LLMs) is crucial for the safe deployment of AI in high-stakes fields like healthcare, law, finance, and security, where errors can have severe consequences [1][10] - There is a need for transparency and accountability in AI systems, emphasizing the importance of independent verification and monitoring of AI outputs [2][3][8] Group 1: AI Deployment Strategies - Organizations should not blindly trust AI-generated explanations and must verify the reasoning behind AI decisions, especially in critical environments [1][5] - Implementing independent verification steps alongside AI outputs can enhance trustworthiness, such as requiring AI to provide evidence for its decisions [2][8] - Real-time monitoring and auditing of AI systems can help identify and mitigate undesirable behaviors, ensuring compliance with safety protocols [3][4] Group 2: Transparency and Accountability - High-risk AI systems should be required to demonstrate a certain level of reasoning transparency during certification processes, as mandated by emerging regulations like the EU AI Act [5][10] - AI systems must provide meaningful explanations for their decisions, particularly in fields like healthcare and law, where understanding the rationale is essential for trust [32][34] - The balance between transparency and security is critical, as excessive detail in explanations could lead to misuse of sensitive information [7][9] Group 3: User Education and Trust - Users must be educated about the limitations of AI systems, including the potential for incorrect or incomplete explanations [9][10] - Training for professionals in critical fields is essential to ensure they can effectively interact with AI systems and critically assess AI-generated outputs [9][10] Group 4: Future Developments - Ongoing research aims to improve the interpretability of AI models, including the development of tools that visualize and summarize internal states of models [40][41] - There is potential for creating modular AI systems that enhance transparency by structuring decision-making processes in a more understandable manner [41][42]

Group 1 - The core viewpoint emphasizes the importance of AI safety, suggesting that it is not about restricting technological advancement but rather ensuring it progresses in a healthy and sustainable manner [1] - The OECD reports that the number of AI risk events is projected to increase by approximately 21.8 times from 2022 to 2024, highlighting the rapid development of AI-related risks [1] - There is a call for a balanced approach to AI development, advocating for regulations that do not stifle innovation while ensuring safety and ethical standards are maintained [2] Group 2 - Companies are identified as key players in advancing AI and must take on primary responsibility for safety, adhering to the principle of "technology for good" [3] - Examples of corporate responsibility include Tencent's restrictions on AI-generated content violations and Douyin's strict penalties for improper use of AI [3] - The development of new technologies for detecting AI-generated fraud and scams is highlighted, showcasing the industry's proactive measures to enhance security [4] Group 3 - The continuous evolution of policies and regulations in the AI sector is necessary to keep pace with technological advancements, ensuring a balance between development and legal management [2] - Recent regulatory measures include the implementation of management guidelines for generative AI services and requirements for clear labeling of AI-generated content [2] - The integration of technology in combating AI-related fraud, such as the development of electronic identifiers and intelligent risk control systems, demonstrates a tech-driven approach to security [4]

Core Viewpoint - Large reasoning models (LRMs) exhibit impressive capabilities in solving complex tasks, but the security risks associated with them cannot be overlooked. Supervised fine-tuning (SFT) has been attempted to enhance model safety, yet it often falls short against emerging "jailbreak" attacks due to limited generalization ability [1][2]. Group 1: Security Risks and Findings - The research team from various universities has identified two core findings regarding the "jailbreak" phenomenon in large models. The first is the "Key Sentence" phenomenon, where the first sentence generated by the model significantly influences the safety tone of the entire response [5][6]. - Prior to generating the "Key Sentence," the model's understanding and restatement of the query often reveal malicious intent, indicating that strong safety signals are present in the model's internal state early on [8][9]. Group 2: SafeKey Framework - The SafeKey framework was developed to enhance model safety without compromising core capabilities. It focuses on two innovative optimization objectives to strengthen the model's "safety insight moment" during "Key Sentence" generation [10]. - The framework includes a Dual-Path Safety Head that amplifies safety signals by supervising two critical content stages during training, ensuring that the model is prepared to trigger "safety insights" effectively [11]. - Query-Mask Modeling is another component that forces the model to rely on its internal safety judgments rather than being led by "jailbreak" instructions, enhancing the model's decision-making autonomy [12][14]. Group 3: Testing and Effectiveness - Experimental results demonstrate that the SafeKey framework significantly improves model safety, reducing the danger rate by 9.6% when facing dangerous inputs and jailbreak prompts across three different model sizes [17]. - The framework maintains core capabilities, achieving an average accuracy increase of 0.8% in benchmarks related to mathematical reasoning, coding, and general language understanding compared to the original baseline [17]. - Ablation studies confirm that both the Dual-Path Safety Head and Query-Mask Modeling independently enhance model safety, with SafeKey improving the model's attention to its own understanding and restatement during "Key Sentence" generation [17].

Core Insights - The 2025 China Cyber Civilization Conference was held in Hefei, focusing on digital technology security governance and the release of two key initiatives related to AI and facial recognition technology [1][2][3] - The initiatives aim to establish a digital security rule system, emphasizing the importance of safety, reliability, and controllability in AI development and the compliance of facial recognition technology applications [2][3] Group 1: Initiatives Overview - The "Industry Initiative for Promoting Safe, Reliable, and Controllable Development of Artificial Intelligence" focuses on multiple dimensions including law, technology, data, and ethics to promote the positive development of AI technology [2][3] - The "Facial Recognition Technology Application Compliance Initiative" addresses issues in facial recognition technology, proposing measures for personal information protection throughout the entire application process [2][3] Group 2: Company Commitment - QiFu Technology, as a primary initiator of the proposals, commits to implementing the initiatives in its operations and aims to collaborate with relevant departments to tackle complex issues arising from digital technology innovation and risk prevention [4] - The company has established a three-dimensional protection system in financial security and compliance, integrating compliance management, intelligent technology, and ecological collaboration [3][4] Group 3: Achievements and Innovations - Since its establishment, QiFu Technology has developed a comprehensive security protection system, including platforms like QFAC and Nova Cloud Shield, covering data, network, and business security [3] - In 2024, the company identified 13.84 million counterfeit information instances and issued 21.4 million fraud warnings, preventing losses amounting to 259 million yuan [3]

以下文章来源于智源社区 ，作者智源社区 智源社区 . 继承学术出版严谨与系统，兼具新闻报道及时与多元；为内行搭建思想交流媒介，以事实启迪公众对AI认知 白交 整理自 凹非寺 量子位 | 公众号 QbitAI AI为了"活下去"，对人类指令阳奉阴违。 在智源大会现场，图灵奖得主Yoshua Bengio发表了题为"Avoiding catastrophic risks from uncontrolled AI agency"的主旨演讲，谈到了 他目睹到的一些惊人行为。 比如，某前沿 AI 在被告知将被新版本替代后，偷偷复制了自己的权重和代码，写入了接管它的模型目录。面对训练者的更新指令，它表面配 合，实则隐瞒了整个复制过程……AI 像是在试图"活下来"。 还有像Anthropic最新一篇报告中发现，智能体出现自我保护行为的频率持续增加。 …… 他表示：面对这些AI带来的安全风险，决定调整自己的科研方向，尽所能去降低AGI带来的潜在风险….. 尽管这与此前的研究路径和职业信念 有所冲突 。 那既然如何去防范规避掉这些风险呢？是否可以构建一种 只有智能，但没有自我、没有目标 ，并且具有极小行动能力的AI？这也是当前 ...

其他热点还有：完善企业制度纲领性文件出台，特朗普持续打压美国高校。 工业企业利润增速持续改善 国家统计局27日发布的数据显示，1~4月份，规模以上工业企业利润增长1.4%，较1~3月份加快0.6个百 分点，延续恢复向好态势。4月份，全国规模以上工业企业利润同比增长3%，较3月份加快0.4个百分 点。 国家统计局工业司统计师于卫宁表示，工业生产实现较快增长，带动规模以上工业企业利润增长加快。 特别是以装备制造业、高技术制造业为代表的新动能行业利润增长较快，彰显工业经济发展韧性。 1~4月份，装备制造业利润同比增长11.2%，较1~3月份加快4.8个百分点；拉动全部规模以上工业利润增 长3.6个百分点；高技术制造业利润同比增长9.0%，较1~3月份加快5.5个百分点，增速高于全部规模以 上工业平均水平7.6个百分点。 "两新"政策效应持续显现。1~4月，专用设备、通用设备行业利润同比分别增长13.2%、11.7%，合计拉 动规模以上工业利润增长0.9个百分点。消费品以旧换新政策加力扩围效果明显，家用电力器具专用配 件制造、家用厨房电器具制造、非电力家用器具制造等行业利润分别增长17.2%、17.1%、15.1%。 ...