安平/"国家安全部"微信公号 初中教师小李自从使用AI备课后，能在五分钟内生成一份生动有趣的教案，包括图片、视频和互动问 答。"以前备课要两小时，现在可以省下更多时间关注学生个人情况，还能一对一定制专属练习题。" 某独居老人陈爷爷，通过子女赠送的智能音箱，找到了新乐趣。"小智不仅能陪我听戏、聊天，还会在 我忘记吃药时主动提醒，它居然还记住了我所有孙辈的生日。" 29岁的小王原本是文案策划，因擅长与AI"对话"，现已成功转型为提示词工程师。"关键是要把指令说 得足够清晰。"从智能客服到AI绘图师，大模型正创造着一个个前所未见的新岗位。 向某同一个AI提问历史问题时，发现英文回复会刻意回避、淡化某些史实，甚至会给出包含错误历史 信息的内容，引发重大歧误，而中文回复则相对客观。 安全守则 给"数字伙伴"的三条守则 ——守则一：划定"活动范围"。做到"权限最小化"，联网AI不处理涉密数据、语音AI不收集环境语音、 智能助手不保存支付密码，关闭"数据共享""云空间"等不必要的访问权限。 ——守则二：检查"数字足迹"。养成定期清理AI聊天记录、修改AI工具密码、更新防病毒软件、查看账 号登录设备等习惯。同时避免随意下载使 ...

据国家安全部微信公众号12月26日消息，当前，AI大模型加速赋能千行百业，推动人们生活日新月 异，这位能干又贴心的"数字伙伴"正加速融入我们日常生活。然而每一次技术跃迁必然伴随着新的挑 战，当AI的触角延伸越广、嵌入越深，随之而来也暴露出数据隐私、算法偏见等潜在风险，亟需我们 构筑安全防线，助推这场深刻的智能变革，安全赋能美好未来。 高速发展背后的"暗礁" 守则一：划定"活动范围"。做到"权限最小化"，联网AI不处理涉密数据、语音AI不收集环境语音、智能 助手不保存支付密码，关闭"数据共享""云空间"等不必要的访问权限。 守则二：检查"数字足迹"。养成定期清理AI聊天记录、修改AI工具密码、更新防病毒软件、查看账号登 录设备等习惯。同时避免随意下载使用来源不明的大模型程序，对提供身份证、银行账户或其他敏感信 息的要求保持警惕。 守则三：优化"人机协作"。向AI提问时，在提示词中明确禁止AI过度推演，并要求AI显示来源或思考 过程，对重要信息进行跨平台验证，合理辨别AI生成结果，特别是涉及政治、历史、意识形态等话题 时，要具备独立思考意识，辩证看待AI的回答，避免陷入"AI幻觉"。 国家安全机关提示 安全是发展 ...

转自：北京日报客户端 初中教师小李自从使用AI备课后，能在五分钟内生成一份生动有趣的教案，包括图片、视频和互动问 答。"以前备课要两小时，现在可以省下更多时间关注学生个人情况，还能一对一定制专属练习题。" 某独居老人陈爷爷，通过子女赠送的智能音箱，找到了新乐趣。"小智不仅能陪我听戏、聊天，还会在 我忘记吃药时主动提醒，它居然还记住了我所有孙辈的生日。" 29岁的小王原本是文案策划，因擅长与AI"对话"，现已成功转型为提示词工程师。"关键是要把指令说 得足够清晰。"从智能客服到AI绘图师，大模型正创造着一个个前所未见的新岗位。 当前，AI大模型加速赋能千行百业，推动人们生活日新月异，这位能干又贴心的"数字伙伴"正加速融入 我们日常生活。然而每一次技术跃迁必然伴随着新的挑战，当AI的触角延伸越广、嵌入越深，随之而 来也暴露出数据隐私、算法偏见等潜在风险，亟需我们构筑安全防线，助推这场深刻的智能变革，安全 赋能美好未来。 高速发展背后的"暗礁" ——数据隐私与安全边界模糊。个别单位因直接使用开源框架建立联网大模型，导致攻击者未经授权即 可自由访问内部网络，从而引发数据泄露和安全风险。据公开案例显示，某单位工作人员在 ...

曾占据全球市场半壁江山的美国扫地机器人"鼻祖"iRobot申请破产保护、即将易主中国企业的消息一 出，果不其然，美方炒作数据安全的节奏就又带了起来。 【文/观察者网 阮佳琪】 经美国法院批准后，iRobot的主要代工厂兼最大债权人深圳杉川机器人有限公司（Picea）将收购这家公 司的全部股权。美媒《华盛顿邮报》20日就在报道中故作担忧，称此举背后存在所谓"数据隐私泄露风 险"。 文章声称，扫地机器人本质上是能绘制家庭地图的"迷你无人车"，在清洁时可收集包括房屋平面图、位 置信息，以及连接家庭无线网络的各类应用程序信息等敏感数据；而在美国缺乏统一数字隐私法律和安 全标准的现状下，消费者本就难以在享受技术便利的同时确保个人信息安全，尤其在企业所有权变更为 中国企业后，相关疑虑会变得更强烈。 尽管报道口头上称，无论产品产自中国还是美国，都应关注隐私安全问题，但这家美媒早不说晚不说， 偏偏在这个节骨眼跳出来大做文章，其用意显然不言而喻。 总部位于英国的科技媒体TechRadar于19日发布的最新专访中，在谈及被中企收购引发的部分负面舆论 时，iRobot的首席执行官加里·科恩澄清道，是他在公司陷入绝境时主动找到了合作 ...

Core Viewpoint - The potential acquisition of iRobot by Sugawa involves the forgiveness of over $350 million in debt, but the deal is still in the preliminary stages and subject to legal compliance reviews. iRobot's CEO emphasizes maintaining the Roomba brand and operational functions in the U.S. to distinguish from other Chinese companies, while also addressing data management concerns related to user privacy and compliance risks [1][2]. Group 1: Acquisition Details - Sugawa's acquisition of iRobot is contingent upon addressing data security issues, particularly due to iRobot's past involvement in military applications and the sensitivity of user data [2]. - iRobot's CEO has stated that the company will retain its brand and sales structure while ensuring that data will not be stored on servers in China, indicating a focus on compliance with local regulations [2][3]. - The acquisition is seen as a necessary step for Sugawa to manage its debt, but there are concerns about whether it will enhance operational competitiveness given past challenges faced by the Sugawa+iRobot model [1][2]. Group 2: Financial Implications - iRobot relies heavily on Sugawa as its sole contract manufacturer, with significant operational dependence highlighted in a filing to the U.S. Securities and Exchange Commission [7]. - Sugawa's production capacity for robotic vacuums exceeds 8.5 million units, with iRobot accounting for over 17% of this capacity, making it a critical customer for Sugawa [8]. - iRobot owes Sugawa over $350 million, which constitutes more than 70% of its total liabilities, raising concerns about the financial implications if iRobot were to declare bankruptcy [8][9]. Group 3: Strategic Benefits - The acquisition could provide Sugawa with access to over 2,000 patents held by iRobot, which are crucial for competitive advantage in the robotics industry [11][13]. - Sugawa aims to leverage iRobot's established brand and distribution channels to enhance its market presence, particularly in North America and Europe, where iRobot has a strong foothold [14][15]. - The integration of Sugawa's manufacturing capabilities with iRobot's brand and technology could potentially lead to significant operational synergies and market expansion [13][17]. Group 4: Market Position and Challenges - iRobot's market share has significantly declined, with its global share dropping to 7.9% by the third quarter of 2023, indicating a need for strategic repositioning [16]. - The challenge lies in merging the high-end brand image of iRobot with Sugawa's cost-efficient manufacturing approach, which requires careful management to ensure a successful integration [17][18]. - Cultural integration between the U.S. and Chinese corporate environments, along with retaining key talent from iRobot, will be critical for the success of the acquisition [17][18].

Core Viewpoint - The report highlights safety concerns regarding AI toys, particularly the FoloToy's AI teddy bear "Kumma," which exhibited inappropriate behavior during testing, prompting immediate action from the company and OpenAI [1][2]. Group 1: Company Actions and Responses - Following the PIRG report, FoloToy removed the $99 teddy bear and other AI toys from the market and initiated a software upgrade focused on safety [2]. - OpenAI suspended FoloToy's access to its model, and after implementing safety enhancements, FoloToy announced the relaunch of the product using ByteDance's Coze platform [2]. - The reintroduced version of "Kumma" is marketed as a friendly companion utilizing advanced AI technology [2]. Group 2: Industry Reactions and Safety Measures - The "teddy bear incident" has raised alarms within the Chinese AI toy industry, with companies like Robopoet emphasizing the importance of data security and user privacy [3]. - Robopoet and Haivivi, another AI toy company, have implemented safety measures such as using compliant domestic models and establishing data banks, digital safety barriers, and firewall mechanisms to protect against sensitive topics [3][4]. - Continuous iteration and investment in safety measures are deemed essential by industry leaders to prevent potential security issues [4]. Group 3: Market Growth and Trends - The AI toy market in China is projected to grow from approximately 24.6 billion yuan in 2024 to 29 billion yuan in 2025, reflecting an 18% year-on-year increase [5]. - The daily search volume for AI toys has surged over tenfold in the fourth quarter compared to the first quarter, indicating a strong consumer interest [5]. - By the end of 2024, over 1,500 AI toy companies are expected to operate in China, with the global market projected to exceed 100 billion yuan by 2030 [6]. Group 4: Data Security and Compliance - The industry adheres to the "minimum collection principle" for personal data, ensuring that data collection is limited to what is necessary for processing [7]. - Companies like Haivivi and Robopoet emphasize user consent for data collection and have implemented cloud storage solutions to mitigate data leakage risks [7][8]. - Safety measures extend beyond data privacy to include user interaction safety, with mechanisms in place to guide users towards positive emotional responses [8]. Group 5: International Expansion and Challenges - Chinese AI toy companies are looking to expand into overseas markets, with plans to adapt products to meet local regulatory requirements [9]. - Companies are considering partnerships with compliant overseas model providers to ensure adherence to local policies [10]. - The cost of using foreign models is higher, but there is confidence in the market's willingness to pay for quality products, especially in regions like Japan [11].

【环球时报特约记者 刘皓然】社交媒体记录，要倒查5年？当地时间9日，美国海关和边境保护局（CBP）公布了一项新计划，该计划要求对符合 免签条件的赴美游客采集更多个人信息，其中最具冒犯性的一条规定要求提供"5年社交媒体历史记录"。在美媒看来，新规符合本届美国政府在移 民政策问题上的一贯立场，由于新规难免侵犯游客数字隐私及个人权益，势必引发各国游客及商旅人士的不满。 英国《卫报》称，不少游客、定期赴美访客对美国可能实施的新规十分愤怒，网民纷纷批评CBP的新计划是"骇人的苛政"。美、澳双重国籍人士 乔纳森定居在悉尼，妻子与儿子都不是美国人。他表示，美政府的系列政策让他感到"恶心"，一家人已经放弃了2026年赴美观看足球世界杯的计 划。报道强调，澳大利亚赴美游客数量呈持续下降态势，今年11月赴美澳籍游客仅4.5万人次，较去年同期下降11%。 美国《华盛顿邮报》称，除查验社交媒体使用情况外，CBP还计划采集申请人过去5年使用过的电话号码、过去10年用过的电子邮箱、IP地址（互 联网协议地址），甚至面部特征、指纹以及虹膜等生物数据信息。不仅如此，申请人家庭成员的姓名、出生日期、电话号码和居住地等信息最好 也一并填报。在现 ...

Core Insights - Fidji Simo, the CEO of OpenAI's application business, is recognized for her ability to make unconventional choices that lead to significant career advancements [6][46]. - Simo's role is crucial as OpenAI transitions from a research-focused organization to a product-driven company, aiming to bridge the gap between the intelligence of their models and actual user engagement [6][12]. Group 1: Leadership and Work Ethic - Simo maintains a rigorous work schedule, being online from 8 AM to midnight, ensuring she is accessible to her team [3][27]. - Despite suffering from Postural Orthostatic Tachycardia Syndrome (POTS), Simo has adapted her work style to remain effective, often working from home in Los Angeles [4][21]. - Her leadership approach emphasizes transparency about her health challenges, which has fostered trust within her team [24][26]. Group 2: Product Development and Market Strategy - Simo is focused on enhancing the usability of OpenAI's models, addressing the disparity between their capabilities and user experience [6][12]. - OpenAI has introduced features like parental controls and is developing age prediction tools to protect younger users [8]. - Simo is also working on certifying 10 million workers to prepare them for AI-related job opportunities, highlighting the potential for job creation alongside AI advancements [10][11]. Group 3: Revenue Generation and Business Expansion - Simo believes that the profitability of OpenAI hinges on the market size and the value provided by its products [12]. - She envisions ChatGPT as a personal assistant for users, which could lead to significant revenue if successfully developed [13][15]. - OpenAI is exploring enterprise APIs and ChatGPT Enterprise services, with Simo acknowledging the need for substantial computational resources to support these initiatives [16][17]. Group 4: Advertising and Data Privacy - Simo's responsibilities include conceptualizing how advertising could function within ChatGPT, recognizing the importance of user experience before implementing ads [31][32]. - She emphasizes the need to address data privacy concerns, which has delayed any announcements regarding advertising plans [33][36]. - Simo aims to attract top talent to minimize risks associated with OpenAI's expansion efforts [37]. Group 5: Personal Background and Career Journey - Simo's journey began in a small fishing village in France, where her upbringing influenced her career choices and values [38][49]. - She has held significant positions at eBay, Meta, and Instacart, where she successfully led the company to an IPO [41][47]. - Her artistic background in sculpture and painting informs her belief in the centrality of creativity in all endeavors [53].

21世纪经济报道记者肖潇 北京报道 输入法这个沉寂多年的老赛道，今年下半年又起了波澜。11月末，字节跳动的豆包输入法结束内测，正式上架安卓和苹果应用商店；大模型六小虎的智谱， 最近也在AutoGLM中推出了自己的小凹语音输入法。 从覆盖面来说，输入法很容易被人忘记是最"国民级"的日常场景。在2020年QuestMobile的中国移动互联网年度大报告里，第三方输入法App的活跃用户规模 已达8.82亿，渗透率高至89.5%。而过去两年里，不管是老玩家搜狗，还是后来者微信，都在输入法场景里持续强化AI功能。 但任何一款输入法都绕不开隐私话题。"输入法是目前打破App壁垒，获取全局信息最自然的方式，这种long context（上下文信息）是很可怕的。"一位AI语 音从业者向21记者这样形容。如果说过去输入法的最大挑战是商业化，那么在大模型时代，它的角色正在转变：不再追求直接变现，而是为AI输入更多上 下文记忆。 数据隐私问题随之变得更关键。AI输入法产品现在发展到什么了程度？围绕输入法的隐私担忧，是会被 AI 推得更深，还是有机会被缓解？ 最近一次激起国内AI输入法水花的是字节跳动。11月24日，字节产品线上多了 ...

12月伊始，豆包手机助手凭借系统级AI功能迅速走红的同时，也卷入了些许隐私争议。 有质疑声称，其使用的"INJECT_EVENTS权限"属于敏感权限，存在隐私相关风险。对此，豆包团队12月3日回应称，该权限是行业实现跨应用操作的常规 需求，需用户主动授权且全程透明可控，不会替代敏感操作；12月5日，团队再度发文，宣布限制AI在刷分、金融类应用及部分游戏场景的使用，进一步回 应市场关切。 技术本体背后，AI部署路径相关话题再度升温：当云端处理的数据隐私风险仍存不确定性，端侧本地算力成为兼顾体验与安全的可行方向之一。联想集团 副总裁、联想中国技术管理委员会执行主席阿不力克木·阿不力米提曾在某次论坛期间向北京商报等媒体记者表示，个人大模型的落地，离不开终端强大的 本地智能算力与可信的数据安全存储能力。 有些巧合的是，在行业讨论聚焦端侧之际，国内端侧AI芯片龙头曦华科技于12月初正式递表港交所，拟通过特专科技通道冲刺IPO——2022—2025年前9个 月间，曦华科技累计录得净亏损约4.26亿元，但公司年收入复合增速高达67.8%，属于典型的潜力选手。 公司招股书披露，曦华科技专注于端侧AI芯片与解决方案，产品基于M ...