央视网消息：据国家安全部公众号消息，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。" ——算法偏见与决策"黑箱"。AI的判断源于其所学习的资料，若训练数据本身存在社会偏见或代表性不足等问题，大模型便可能放大歧视。测试显示，某 些AI会系统性偏向西方视角。当研究人员用中英文向某同一个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 ...

人民网上海12月25日电 12月25日，第七届人民网内容科技论坛在上海模速空间隆重举行。会上， 《中国内容科技十大进展（2021-2025）》报告正式发布。报告系统梳理了"十四五"期间我国内容科技 领域取得的一系列突破性成果，勾勒出从"工具赋能"迈向"生态重构"的清晰发展脉络，展现了智能化、 垂直化、体系化的鲜明发展态势。 报告指出，内容科技作为推动内容生产、传播、审核及治理全链条智能化升级的关键力量，正深刻 重塑传媒生态与数字社会面貌。此次发布的十大进展，集中体现了我国在该领域的创新活力与治理智 慧。 进展一：国产通用大模型崛起与媒体融合应用。 2023年以来，以"文心一言""通义千问"等为代表的国产大模型集中上线，DeepSeek等开源模型以其 高性能、低成本等特点备受关注。大模型积极向媒体等垂直领域深度渗透，催生人民网"写易"、人民智 媒大模型、新华社MediaGPT、"央视听大模型"等行业专用模型，成为推动内容生产、交互与服务智能 化升级的新型基础设施，显著拓展了媒体融合的深度与广度。 进展六：智能修复与数字活化助力文化遗产传承创新。 2020年以来，基于GAN和扩散模型的影像修复技术日趋成熟，可实现 ...

深度伪造指用人工智能(AI)技术制作看起来真实度很高的视频、图片或音频片段，这种技术可将色情影 像中的人物改变成另外一人的样貌。这类犯罪事件大多针对女性和未成年人。 日本警方17日说，警方今年前9个月掌握涉及18岁以下未成年人的深度伪造色情内容案件中，约半数出 自受害者同学之手。 日本共同社18日援引日本警察厅数据报道，日本警方今年前9个月处理的79起此类案件中，逾八成受害 者为初中生或高中生，其中约半数案件的作案人与受害者在同一所学校就读。用于生成深度伪造色情内 容的照片，通常取自毕业纪念册。 这是日本警察厅首次就此类案件发布详细数据。上述案件受害者中，初中生占比约52%，高中生占比约 32%，小学生占比约5%。约53%案件中的深度伪造图像由受害者的同学制作或传播；约6%的案件作案 人是受害者通过社交媒体结识的人，另有6%案件由受害者所在学校教职员工或其他学校的学生实施。 在日本，未经当事人同意制作并传播其色情图像构成侵权，涉案者还可能面临诉讼。上述79起案件中， 日本警方对4起以涉嫌诽谤罪等为由展开刑事调查，另有6起对实施者给予行为矫正指导。 日本警察厅说，上述案件中，约18%涉及生成式AI制作的照片。虽 ...

Core Viewpoint - The rise of "AI Slop" content, characterized by low-quality, repetitive, and meaningless material generated by AI tools, is increasingly prevalent on the internet, particularly on social media platforms [1][2][4]. Group 1: Definition and Characteristics of "AI Slop" - "AI Slop" refers to low-quality content produced by AI tools, including text, images, and videos, often found on social media and content farms [2][3]. - The term "Slop" originally described cheap and low-nutrition items, and its modern usage highlights the poor quality of AI-generated content [2]. - Unlike "deepfakes" or "AI hallucinations," which have specific deceptive intents or technical errors, "AI Slop" is produced without regard for accuracy or logic, leading to a flood of meaningless content [3]. Group 2: Causes of Proliferation - The proliferation of "AI Slop" is driven by the increasing power and low cost of AI technology, enabling rapid content generation that prioritizes clicks and ad revenue over quality [4]. - New AI tools like ChatGPT, Gemini, and Sora allow for quick production of readable text, images, and videos, leading to the rise of content farms that prioritize quantity over quality [4]. - Algorithms on social media platforms often favor engagement metrics over content quality, further encouraging the spread of "AI Slop" [4]. Group 3: Consequences of "AI Slop" - The overwhelming presence of "AI Slop" can obscure credible sources in search results, blurring the line between truth and fiction [5][6]. - As misinformation spreads more rapidly in an environment where distinguishing fact from fiction becomes challenging, the trust crisis in information sources intensifies [6]. Group 4: Potential Solutions - Some companies, like Spotify, are beginning to label AI-generated content and adjust algorithms to reduce the visibility of low-quality material [7]. - The C2PA (Coalition for Content Provenance and Authenticity) standard aims to embed metadata in digital files to trace their origins, helping to differentiate between human-created and AI-generated content [7]. - The most effective defense against "AI Slop" lies in individual responsibility, encouraging users to verify sources and support genuine creators [7][8].

◎科技日报记者 刘 霞 "Slop"原指"猪食""泔水"等廉价、粗糙、缺乏营养之物。如今，借由AI技术的浪潮，一些如同"Slop"的 垃圾内容正在互联网上肆意蔓延。 "AI垃圾"特指由AI工具生成的大量劣质、重复或无意义的文字、图像或视频，常见于社交媒体和自动化 内容农场。 美国科技网站CNET在10月28日的报道中提到，如今社交平台上"AI垃圾"五花八门：OpenAI的Sora可让 人几秒内生成滑稽荒诞的视频；LinkedIn上满是由AI包装的"专家箴言"，例如"有时领导力就是保持沉 默的能力"；谷歌搜索结果也宛如AI杂货铺，竟会出现"姜黄能治愈心碎"这类无稽之谈。 "AI垃圾"与"深度伪造"或"AI幻觉"虽有重叠，却并不相同，其区别在于意图与质量。 "深度伪造"指利用AI伪造或篡改影音，目的在于欺骗，从虚假政治演讲到诈骗语音皆属此类，其关键在 于以假乱真。"AI幻觉"则属技术错误，聊天机器人可能引用不存在的研究，或编造法律案例，本质是模 型在预测词语时出现了偏差。而"AI垃圾"范围更广，也更随意：当人们用AI批量生产内容却疏于核对准 确性与逻辑时，便会产生此类垃圾。它们堵塞信息渠道，抬高广告收益，用重复无 ...

"假冒伪劣"信息充斥网络 如今的互联网上，看似信息海量，但也充斥着大量单调、重复且缺乏质量的内容。 美国《纽约时报》网站在12月8日的报道中指出，当前网络，尤其社交平台正泛滥一种被称为"AI垃 圾"（AI Slop）的内容。英国《新科学家》网站10日也发表文章称，今年，许多人感觉仿佛置身于一堆 华而不实的"AI垃圾"中。英国《经济学人》杂志更是将"Slop"一词选为2025年度词汇。这类错漏百出、 古怪甚至令人尴尬的内容遍布各平台，也在悄然侵蚀着人们的思想。 "Slop"原指"猪食""泔水"等廉价、粗糙、缺乏营养之物。如今，借由AI技术的浪潮，一些如同"Slop"的 垃圾内容正在互联网上肆意蔓延。 "AI垃圾"特指由AI工具生成的大量劣质、重复或无意义的文字、图像或视频，常见于社交媒体和自动化 内容农场。 美国科技网站CNET在10月28日的报道中提到，如今社交平台上"AI垃圾"五花八门：OpenAI的Sora可让 人几秒内生成滑稽荒诞的视频；LinkedIn上满是由AI包装的"专家箴言"，例如"有时领导力就是保持沉 默的能力"；谷歌搜索结果也宛如AI杂货铺，竟会出现"姜黄能治愈心碎"这类无稽之谈。 "AI垃 ...

典型诈骗场景主要包括：合成子女、父母或好友的影像，在视频通话中以"急需用钱"为由要求转账；针 对企业财务人员或员工，冒充公司领导通过视频会议下达紧急支付指令；甚至伪造执法人员视频，以涉 嫌违法为由要求消费者将资金转移至所谓"安全账户"进行核查。 面对这类利用生物特征进行欺骗的新型犯罪，山西省消协提示消费者，关键在于建立"多重验证"意识， 不可单纯信赖眼见耳闻之象，并应谨记"三不二要"防范原则。 "三不"即不轻信、不转账、不泄露。对于任何通过非见面方式提出的转账汇款要求，即便看到真人视频 或听到熟悉声音，也应高度警惕，尤其要提防对方制造紧急氛围阻碍核实。在未通过其他可靠途径百分 之百确认对方身份前，坚决不进行任何转账操作。同时，需注重个人信息保护，审慎在社交媒体分享清 晰的正面视频、语音，避免敏感信息过度公开。 "二要"即要核实、要警惕。若遇"亲友"通过视频借钱，务必挂断后使用自行存储的联系方式回拨核实， 或通过共同亲友交叉确认；可与家人、挚友预先约定资金往来暗语。此外，需保持观察，AI合成内容 可能存在人物表情僵硬、眨眼不自然、口型对不上、声音有电子音或断句不合理等细微破绽。消费者需 牢记，公检法等机关绝不会 ...

Core Insights - Deepfake technology has evolved from a conceptual threat to a tangible business risk, as evidenced by incidents like the AI-generated image of an explosion at the Pentagon that caused a significant drop in the S&P 500 index, erasing billions in market value [1] - The deepfake market is projected to grow from $75 billion in 2023 to $385 billion by 2032, highlighting the urgent need for businesses to upgrade their defense systems against misinformation [1] Group 1: Threats of Deepfake Technology - Deepfake attacks pose a direct threat to corporate assets, as demonstrated by the case of engineering giant Arup, which lost 200 million HKD due to a sophisticated scam involving AI-generated executive voices [2] - The advertising giant WPP also faced a similar attack, where scammers attempted to replicate the CEO's voice and appearance to deceive employees into transferring funds [2] Group 2: Operational Challenges - Companies that rely on facial recognition technology will need to replace their security systems by 2026, as existing solutions are ineffective against deepfake threats [3] - The addition of verification measures, such as digital watermarks and live detection, increases operational costs and complicates decision-making processes [3] Group 3: Trust Crisis - The rise of deepfake technology is leading to a "trust tax," which includes both direct security investments and indirect costs arising from widespread skepticism in digital communications [4] - The erosion of trust in digital interactions complicates business collaborations, as every communication may require additional verification [4] Group 4: Strategic Solutions - Companies should focus on establishing credible verification methods, such as digital signatures and watermarks on sensitive documents, to mitigate the risks posed by deepfakes [5][6] - Creating a public verification center can provide stakeholders with authoritative information and enhance trust in corporate communications [7] - Training employees to recognize deepfakes and incorporating simulation exercises into onboarding processes can improve organizational resilience [8] - Investing in real-time media tampering detection systems is essential for embedding verification capabilities into core business processes [9] Group 5: Navigating the Crisis - The rise of AI-generated content is accelerating the collapse of shared realities, making it crucial for companies to anchor their communications in verifiable facts [10] - Organizations that prioritize building trustworthy systems will not only withstand the challenges posed by deepfakes but also emerge as leaders in their industries during chaotic times [10]

Core Insights - The South Korean National Police Agency reported a significant increase in arrests related to online sexual crimes, with over 3,000 suspects apprehended in the past year, marking a 47.8% increase compared to the previous year [1] Summary by Categories Crime Statistics - From November 2023 to October 2024, the police solved 3,411 cases of online sexual crimes and arrested 3,557 suspects, with 221 formally detained [1] - Cases involving deepfake technology accounted for the highest proportion at 35.2%, followed by child or adolescent pornography at 34.3%, and illegal filming at 19.4% [1] Demographics of Suspects - The majority of suspects, totaling 1,761, were teenagers aged 10 to 19 [1] - The increase in the number of apprehended suspects is attributed to both the rise in deepfake-related cases and intensified law enforcement efforts [1] Public Awareness and Education - A significant number of South Korean students lack awareness of the dangers associated with deepfake technology in sexual crimes, with 62.2% of middle school students and 47.7% of high school students considering deepfakes as mere "pranks" [1]

Core Viewpoint - The emergence of AI-generated fake invoices poses a significant threat to financial trust, as these sophisticated tools make it easier for individuals to commit fraud [5][6]. Group 1 - Traditional methods of fraud involved basic tools like photocopiers and correction fluid, but advancements in technology have led to more sophisticated techniques [6]. - AI-generated fake invoices are now equipped with realistic trademarks, addresses, and details, even simulating wear and tear like creases and coffee stains [6]. - The potential for "deepfakes"—manipulated videos or audio that can make public figures appear to say things they never said—raises concerns about their use in political and financial fraud [6].