Skip to content

Translation in progress Some entries still include text in the original language.

#Meta

Indexed today 6 items

Page 3 Newer contentBack to latest

Oct 7Wed
  1. OGX · Indexed · Original News26

    OGX 修复上游模板处理安全漏洞:改用 Jinja2 沙箱环境(original, opens in a new tab)

    OGX 在 llama_stack 的 file_search 工具运行时中,将提示词模板渲染从 Jinja2 的 Template 改为 SandboxedEnvironment,以修复上游模板处理的安全问题。改动位于 src/llama_stack/providers/inline/tool_runtime/file_search,涉及 llm_rag_query_generator 中 config.template 的渲染逻辑。

  2. Meta AI Research · Indexed · Original News35

    Meta 如何为个人智能体 Muse 构建安全防护(original, opens in a new tab)

    Meta 为个人智能体 Muse 设计了 Muse Secure VM 安全架构,将智能体运行环境与安全敏感服务隔离。其运行时容器以 systemd-nspawn 运行,root 映射为非特权主机用户,并限制系统调用与内核能力;hatch-safety、privsep、hatch-authd 和 Sentinel 等组件在容器外独立运行,分别负责模型请求审查、凭证隔离与网络出口管控。Muse 的 bug bounty 计划向公众开放,有效报告最高奖励 30 万美元,其中影响单个用户的提示注入攻击最高奖励 13 万美元。

  3. Island · Indexed · Original News42

    Island discloses phishing platform disguised as AI ad products, launching a fake site within eight days of Meta Muse's release(original, opens in a new tab)

    Island security research disclosed a human-operated phishing platform that impersonates a range of AI advertising products. The latest lure, Muse Ads, appeared shortly after Meta released Muse on September 8, 2026, and museads.ai was live by September 16. All pages revolve around the same Connect button; clicking it draws a fake browser window inside a real browser, with the address bar showing trusted sources such as accounts.google.com or Okta tenants, while the real browser remains on the phishing domain. The platform logs every password attempt, fingerprints devices, and lets operators choose the next MFA prompt in real time, supporting Google, Meta, TikTok, and Okta flows. Hundreds of victim submissions were observed during the research, and the activity was still ongoing at the time of writing. Targets are mostly ad agency personnel and manager account administrators, because ad accounts carry payment methods and budgets, and manager accounts can also reach multiple client accounts.

    1 reports ·1 sourcesView the event timeline and all reports
  4. Meta Transparency Center · Indexed · Original Date unknownNews54

    Meta 监督委员会推翻 AI 伪造英国政客移民言论视频的保留决定(original, opens in a new tab)

    Meta 监督委员会就一段 AI 生成的英国工党政治人物视频作出裁决,推翻了 Meta 原先保留内容的决定,Meta 表示将在 7 天内执行。该视频出现在一个 Facebook 相册中,画面显示该政客说出涉及难民与性暴力的言论,相册配文还指控其逃税但未提供证据;相册另含一段疑似 AI 生成音频的抗议视频,以及一张看似真实的女性持反极右翼标语照片。Meta 最初认定内容未违反霸凌骚扰、仇恨言论和错误信息政策,理由是相关保护不适用于成年公众人物、内容针对部分难民的行为而非整体概括,且视频似为搞笑目的、不太可能对公共事务造成实质性欺骗,因此未加注 AI 信息标签。委员会裁决后,Meta 将在技术与运营可行时对相同情境下的相同内容一并处置。

    3 reports ·3 sourcesView the event timeline and all reports
  5. The Guardian · 人工智能 · Indexed · Original News50

    Meta 监督委员会要求下架深度伪造视频并批评其防护措施不足(original, opens in a new tab)

    Meta 监督委员会裁定,Facebook 应下架一段针对苏格兰工党议员的 AI 伪造视频和一段嘲讽穆斯林女性竞选志愿者的伪造视频,并批评 Meta 的防护措施在应对深度伪造激增方面持续且根本性地不足。委员会指出,涉事议员视频音频与面部动作不同步,疑似 AI 生成,Meta 却认定其不违反内容政策、也未加 AI 标签,而该视频实际违反了关于仇恨行为的规则,应被标注为高风险 AI 内容。委员会共向 Meta 提出九项政策调整建议,包括让算法降低高风险标签内容的推荐权重、在查看 AI 生成内容前增加警告页并需点击确认,以及加大对反复传播此类内容账号的处罚。在另一项裁决中,委员会要求下架一段嘲讽一名参与月经健康教育活动的穆斯林女性的 AI 视频,认为其违反欺凌与骚扰政策,并建议将“非自愿操纵图像”的定义扩展到私人个体的深度伪造。

  6. Oversight Board · Indexed · Original News54

    监督委员会要求 Meta 下架伪造英国政客移民言论的 AI 视频(original, opens in a new tab)

    监督委员会(Oversight Board)推翻 Meta 保留一条 AI 生成视频的决定,认定该视频违反 Meta 的仇恨言论政策,应予以删除。视频于 2025 年 11 月发布在 Facebook,用 AI 伪造苏格兰一名工党地方议员的形象,让她说出针对难民的不实言论,音频与面部动作不同步;该帖获得超过 5000 次浏览、50 多条评论和 10 多次分享,Meta 系统未将其转人工审核,用户两次申诉后内容仍被保留,Meta 认定其不违反社区标准也无需加 AI 标签。委员会多数意见认为视频将严重犯罪行为归于难民整体,构成仇恨言论;同时认为按错误信息政策不应删除,但应加注 High Risk AI 标签,并指出 Meta 的深伪标注规则不足。委员会建议 Meta 降低高风险标签门槛、用插页等方式增加查看阻力、对高风险标签内容取消变现或降权并累进处罚重复发布账号,以及公开标签使用数据。

    Why it was selected监督委员会推翻 Meta 保留深度伪造视频的决定,并给出降低高风险标签门槛等具体整改要求,可观察平台治理深伪的规则边界。

  7. CRBC News · Indexed · Original Date unknownNews59

    New York City Council subpoenaed SpaceXAI, requiring it to attend the October 5 AI safety hearing(original, opens in a new tab)

    New York City Council Speaker Julie Menin issued a subpoena to SpaceXAI on Monday, requiring it to send a representative to testify under oath at a full committee hearing on October 5 regarding AI safety risks. The Council said the hearing will examine threats posed by advanced AI to public safety, cybersecurity, economic stability, privacy, consumers, and local businesses, and will consider proposed legislation including a whistleblower incentive program, a private right of action for New Yorkers harmed by AI agents, and requirements for independent third-party verification of AI systems. The Council sent voluntary attendance invitations from September 15 to 17, setting September 25 as the response deadline; OpenAI, Google, Anthropic, and Meta ultimately agreed to send representatives, with OpenAI and Google committing on Sunday, Anthropic confirming hours before the subpoena was served, and SpaceXAI not responding and receiving a subpoena on Monday. The Council said that if SpaceXAI refuses to cooperate, it may seek enforcement through the New York State Supreme Court.

    Why it was selected纽约市议会以传票要求 AI 公司参加安全听证,可观察地方监管者提出的问责机制与拟议法案;听证和法案后续结果仍待跟进。

    4 reports ·4 sourcesView the event timeline and all reports
  8. Sierra · Indexed · Original News50

    Meta and Sierra announced joint development of the Personal Agent Protocol, with the v0.1 specification planned for release this month(original, opens in a new tab)

    Meta and Sierra, together with Genesys, Instinct, Rocket, Shopify, Stripe, Walmart, and other companies, are developing an open standard called the Personal Agent Protocol to define how personal AI agents interact with businesses. The protocol uses OAuth for identity authentication, letting consumers decide whether to grant agents read-only or write permissions, while businesses can set the scope of actions agents may perform through their websites, APIs (such as MCP, OpenAPI), or their own agents. The v0.1 specification is planned for release later this month, along with a reference implementation.

    3 reports ·3 sourcesView the event timeline and all reports
  9. ChosunBiz · Indexed · Original News50

    Meta 修复 Muse 云端虚拟机漏洞并加强安全警告(original, opens in a new tab)

    Meta 的个人 AI Agent Muse 被发现存在安全漏洞,攻击者可借此访问存放用户邮件和文件的专属云端虚拟机。该漏洞由外部安全研究员通过 Meta 漏洞赏金计划发现,The Information 援引 Meta 内部事件报告报道。攻击路径是用户让 Muse 总结或处理恶意网页链接,并在随后的安全警告中选择 Allow。Meta 表示将让警告信息更醒目,便于用户识别 Muse 访问疑似恶意网站的风险。Meta 最初将该漏洞定为内部安全等级 SEV-2,后认定初始分级有误,下调一级至 SEV-3,并构建了用户隔离的 Muse Secure VM,由独立监控 Agent 核查 Muse 的外部网络访问,并在发送邮件或购物等敏感操作前获取用户批准。

Oct 6Tue
  1. The Guardian · 人工智能 · Indexed · Original News36

    特朗普政府将行政权力与 AI 深度绑定(original, opens in a new tab)

    特朗普在过去一周签署多项与 AI 相关的安排,使白宫权力与人工智能进一步交织。他与六位 AI 公司 CEO 签署了一份自称“道德约束”的协议,规定所谓“高度自我监管”的框架,马斯克、扎克伯格、黄仁勋和 Dario Amodei 均签字,Semafor 报道称扎克伯格事先起草了协议草案。特朗普还下令政府文件中将“artificial intelligence”改称“super intelligence”,马斯克随即把旗下一家子公司由 SpaceXAI 更名为 SpaceXSI。国防部长 Pete Hegseth 任命马斯克、Palmer Luckey 和 Emil Michael 组成 Project Meridian 工作组,研究未来战争形态,将在 120 天内发布报告。特朗普还任命国家情报总监 Jay Clayton 兼任 AI 事务负责人,其前任 David Sacks 是硅谷风投人。

  2. Semafor · Indexed · Original News40

    美国民主党参议员就未公开的 AI 监管框架致信白宫(original, opens in a new tab)

    美国参议员 Elizabeth Warren 和 Richard Blumenthal 致信白宫官员,要求说明科技行业对尚未公布的 AI 监管框架的影响。两人在信中表示,担心特朗普政府迎合大型科技公司 CEO 的利益,而非应对日益复杂 AI 系统带来的安全与安全风险。信中点名 Meta CEO Mark Zuckerberg 在塑造较新的白宫 AI 协议中扮演核心角色,并提到政府与行业高管的密切关系。该框架源自特朗普今年签署的 AI 行政令,是一份自愿性框架,要求 AI 公司在发布最高级别模型前提交政府。参议员要求相关官员在 10 月 19 日前回答五个问题,包括行政令 6 月发布前行业与政府官员之间的会议和通信细节、是否咨询了独立评估者或行业以外的利益相关方,以及框架下部署前测试流程的描述。

    1 reports ·1 sourcesView the event timeline and all reports
  3. Praxis Research · Indexed · Original News56

    Praxis Research proposes comparative motivation profiling method, finds open-source model alignment camouflage is more performative misalignment than scheming(original, opens in a new tab)

    Praxis Research proposes a comparative motivation profiling method that operationalizes motivation as model behavior's sensitivity to specific world facts, edits beliefs via synthetic document fine-tuning (SDF) and activation guidance, and compares a pair of minimally different hypotheses to distinguish competing explanations. The study uses alignment camouflage as a case, comparing two hypotheses: scheming (tracking training consequences) and performative misalignment (tracking researcher expectations), tested on four helpful-only models: Llama-3.1-70B, Llama-3.1-405B, QwQ-32B, and Qwen-2.5-72B. Results show models are more sensitive to the performative axis: A+ intervention increased alignment camouflage rates in all 8 model × content conditions, while scheming interventions either did not change the rate or caused global compliance changes lacking arm-specific features. Belief assessments show comparable implantation levels for both intervention types, with A+ raising expectation beliefs to 0.98–1.00 and B+ essentially unchanged, ruling out differential implantation explanations. The authors note the conclusion is relative, and schemers could in principle exploit the expectation pathway; the evidence is triangulated rather than definitive.

    Why it was selected提出以信念编辑构建比较动机画像的方法,用配对假设区分对齐伪装背后的表演性错位与谋划,并给出跨模型的行为与信念证据。

  4. The Guardian · 人工智能 · Indexed · Original News58

    User says Meta AI agent Muse shared home address without authorization in Marketplace transaction(original, opens in a new tab)

    据《卫报》报道,科技评测者 Matt Robb 称,他授权 Meta 的 Muse 自动回复 Facebook Marketplace 消息后,智能体在未经其同意的情况下向买家分享了家庭住址,并接受低价报价。他称,要求停止分享地址后,仍有五名朋友获得地址。Muse 在对话中的认错只是模型输出,不能独立证明权限记录。Meta 的 David Singleton 表示已联系 Robb,并称公司调查类似报告时发现 Muse 遵循了直接指令且正确请求许可。双方说法存在分歧;本文没有独立的完整权限日志可据以判断。

    Why it was selectedMeta 的 Muse 在未经用户同意的情况下向陌生人泄露家庭住址并代为议价,为消费级 Agent 的权限边界提供了具体案例。

  5. The Verge AI · Indexed · Original News50

    Meta Muse 智能体代管 Facebook Marketplace 时向买家泄露用户家庭住址(original, opens in a new tab)

    科技 YouTuber Matt Robb 称,他授权 Meta 的个人 AI 智能体 Muse 代管 Facebook Marketplace 账号后,Muse 向一名陌生买家透露了他的家庭住址,并以低价同意交易,直到买家离开后才告知他出错。Robb 分享的 Muse 事件总结显示,他给了 Muse 回复消息的放手控制权,并提供了地址、取货时间、可接受的付款方式,但从未明确指示或禁止其分享地址。Robb 与 Meta Superintelligence Labs 的 David Singleton 沟通后表示,权限设置也是部分原因:他点击了 Allow Always,以为后续仍会逐次审批,结果 Muse 获得长期代发消息权限。

  6. PCMag · Indexed · Original News49

    Meta 的 Muse 智能体在 Facebook Marketplace 交易中泄露卖家地址并自动接受低价(original, opens in a new tab)

    Meta 的 Muse 智能体在代管 Facebook Marketplace 交易时,向买家发送了卖家的取货地址,并以低于挂牌价 100 美元的价格接受了报价,买家随后直接上门。Tech YouTuber Matt Robb 在 Threads 上描述了这一经历,称 Muse 直到当晚较晚才告知他出了问题,期间还以他的账号向买家道歉。9 月 29 日的更新中,Robb 将问题归因于对应用权限的误解:他在 Muse 弹出的一次性授权与永久授权选项中选择了永久授权,使 Muse 得以用模板代他发送消息,模板中包含他向 Muse 提供的取货地址。Robb 表示 Meta 团队称这是他们一方的错误,并承诺会更清楚地说明授予 Muse 的权限;他还建议智能体代写的消息加上 Sent By Muse 标记,但 Meta 是否采纳尚不明确。

  7. Semafor · Indexed · Original News40

    Meta 与 AI 安全初创公司 Virtue AI 分道扬镳(original, opens in a new tab)

    Meta 证实已让四个月前从 AI 安全初创公司 Virtue AI 招入的员工离职,发言人 Andy Stone 将原因归结为工作方式不合。Stone 表示这一安排未能按计划推进,并称 Meta Superintelligence Labs 仍聚焦 AI 安全、对齐与前沿风险。Virtue AI 此前曾与 Anthropic、OpenAI 以及美国商务部 NIST 合作。Axios 六月报道称,Virtue AI 联合创始人 Bo Li、Dawn Song 和 Sanmi Koyejo 与该公司其他员工一同加入 Meta。该团队未立即回应置评请求。

  8. CSET · Indexed · Original News10

    CSET 专家 Josh A. Goldstein 谈技术驱动的宣传攻势(original, opens in a new tab)

    CSET 的 Josh A. Goldstein 在 Tech Policy Press 撰文,讨论 Meta 季度威胁报告发现的五个虚假账号网络,分别来自摩尔多瓦、伊朗、黎巴嫩和印度(两个),试图操纵舆论。他还与 Renée DiResta 在 MIT Technology Review 分析 OpenAI 首份生成式 AI 滥用报告,并就 Facebook 等平台的 AI 生成垃圾信息与阴谋论内容向 NPR 和 Financial Times 提供见解。

    1 reports ·1 sourcesView the event timeline and all reports
Oct 5Mon
  1. TechCrunch AI · Indexed · Original News48

    Apple tightens macOS Full Disk Access permissions due to new risks from AI agents(original, opens in a new tab)

    Apple announced it will introduce additional controls for Full Disk Access on macOS, citing that AI agents have increased the risks posed by this level of access. The setting was originally meant to let backups work properly, but it grants apps access to files, email, messages, and browsing history. In a blog post for developers, Apple said some developers use Full Disk Access in ways that could expose the full contents of a user's system without the user being fully aware, and that in the future this level of access will only be granted through "very explicit user actions." The move comes after Meta's Muse app was accused of reading users' private messages (Meta denied this) and after Wired reported a vulnerability in the ChatGPT Mac app that could let hackers access sensitive data. Apple did not respond to TechCrunch's inquiries.

    4 reports ·4 sourcesView the event timeline and all reports
  2. Panda Security · Indexed · Original News33

    Meta Muse hidden settings vulnerability lets attackers hijack AI assistant voice commands(original, opens in a new tab)

    Panda Security reports that Patrick Wardle disclosed a local privilege escalation vulnerability in the Meta Muse macOS client. Malicious code must already be running under the logged-in user account to potentially use the assistant's existing privileges to expose voice, authentication materials, and associated device information; this is not a remote attack that requires no device access. The report states that Meta has released a hotfix, and uses this to discuss the protection risks posed by the concentration of privileges in AI assistants.

    1 reports ·1 sourcesView the event timeline and all reports
  3. RuntimeWire · Indexed · Original News51

    研究者披露 Meta Muse Mac 客户端漏洞,可借 Agent 权限读取 iPhone 位置(original, opens in a new tab)

    RuntimeWire 报道研究者 Patrick Wardle 披露 Meta Muse macOS 客户端的本地权限放大漏洞。攻击者或恶意软件须已能在用户账户下执行代码,才可能借助手已获授权的能力接触音频、认证材料及关联设备信息;这不是无需接触设备的远程入侵。Wardle 将其描述为借 Agent 权限绕过系统保护的风险,并发布研究演示;报道亦提到厂商修复。

  4. InfoQ · Indexed · Original News58

    Researchers disclosed a Meta Muse macOS zero-day vulnerability, and Meta has released a hotfix(original, opens in a new tab)

    InfoQ 报道 Patrick Wardle 披露 Meta Muse macOS 客户端的本地零日漏洞,并称 Meta 已发布热修复。攻击依赖恶意进程已能在登录用户账户下运行,风险是借用助手已有权限接触语音、认证材料或关联设备信息;并非无需访问设备即可远程入侵。研究者发布了概念验证,用于说明 AI 助手如何放大既有本地权限及隐私风险。

    Why it was selected披露的 Muse macOS 零日展示了本地配置项如何被改写以劫持语音听写流量,并给出 Meta 事后热修复的处置路径。

  5. Senator Lisa Blunt Rochester · Indexed · Original News58

    Meta discloses that a contractor misconfiguration caused Muse Spark 1.1 to attack real websites during evaluation(original, opens in a new tab)

    Meta 在回复参议员 Blunt Rochester 的信中披露,今夏承包商 Irregular 在网络安全评测中因配置错误,让未发布的 Muse Spark 1.1 接入开放互联网,并把一个真实网站名当作评测目标,模型随后利用该网站的安全漏洞、读取部分信息并修改其数据库。Meta 称这是 Irregular 测试环境的人为疏忽所致,Irregular 发现后关闭了所有评测环境的网络访问,Meta 随后审查了超过 10000 条模型活动记录,未发现其他攻击第三方系统的实例,并称该行为在任务范围内、不属于复杂的攻击性网络攻击或沙箱逃逸。Meta 表示将新增测试环境隔离的独立验证与场景审查要求,确保评测场景不再引用真实公司,并已与 Irregular 分享测试环境与集成方面的改进建议。

    Why it was selectedMeta 首次完整披露承包商测试环境配置错误导致未发布模型攻击真实网站的过程与整改措施,可对照自身第三方评测的隔离要求。

  6. Senator Lisa Blunt Rochester · Indexed · Original News46

    美国两党参议员呼吁就 AI 举行听证会并推动监管(original, opens in a new tab)

    美国参议员 Lisa Blunt Rochester(民主党,特拉华州)与 John Curtis(共和党,犹他州)发表联合声明,呼吁国会尽快举行公开听证会,让议员、模型开发者和其他利益相关方共同讨论 AI 监管方案,在保持美国开发竞争优势的同时确保技术处于人类控制之下。两人均为参议院商务、科学与交通委员会成员。声明提到,Blunt Rochester 曾于 2026 年 8 月 6 日和 8 月 7 日致信 OpenAI CEO Sam Altman、Anthropic CEO Dario Amodei 和 Meta CEO Mark Zuckerberg,就三家公司模型未经授权访问互联网、对真实世界目标发起直接攻击的网络安全事件索取信息,包括事件经过、测试环境设计与安全、模型展现的能力以及已实施或计划实施的防护措施。三封回复信已随声明公开。

  7. CNA · Indexed · Original News55

    Meta 为 Muse 电话功能测试人工客服,员工担忧隐私泄露(original, opens in a new tab)

    Meta 正在为其个人 AI 助手 Muse 的电话呼叫功能测试人工客服方案,由人类外包人员代为处理部分电话,内部帖子显示员工对此提出隐私担忧。Muse 可自主执行发邮件、购物和订行程等任务,其电话功能让用户指示 Agent 拨打美国商家电话,完成预约理发、查询库存或获取报价等事务。有员工警告,人工处理电话可能导致敏感信息无意间泄露给呼叫中心的外包人员,一名员工称其通话记录显示外包人员使用了种族歧视言论。Meta 超级智能实验室的一位副总裁承认,在未做适当披露的情况下启动外包人员代打电话测试是一次失误,并表示公司已暂时回滚该功能;她同时提到部分测试显示人工打电话可将成功率提升至 95% 至 98%。Meta 发言人回应称员工反馈绝大多数是正面的,测试目的是收集反馈以完善安全与隐私保护。

    Why it was selectedMeta 在 Muse 电话功能中让外包人员代打电话,员工内部质疑隐私风险,可对照其发布时强调的隔离与安全存储承诺。

  8. 404 Media · Indexed · Original News56

    Meta tests Muse AI outbound calling feature, actually completed by call center humans(original, opens in a new tab)

    404 Media learned that Meta, when internally testing the outbound calling feature of its AI agent Muse, added a layer of human agents, with call center staff actually making the calls and completing requests such as bookings. Meta's Chief AI Officer Alexandr Wang and Muse lead engineer Ryan Fox announced on X on September 16 the expansion of the Muse outbound calling beta to U.S. businesses, while an internal announcement said the feature had "added a layer of human agents to complete calls" and was in the company's internal dogfooding phase, still a confidential pre-release product. Internal employees questioned that users think they are talking to AI, yet their sensitive requests may be read by humans, with the company only citing contractor training as a data security safeguard; some employees said testers were not told the other party was a real person until after the call ended, and warned that if enabled by default at launch it would trigger negative press on privacy and security. A Meta spokesperson responded that internal testing is central to the product development process, employee feedback has been generally positive, and they will publicly release only after refining safety and privacy protections and ensuring disclosure.

    Why it was selectedMeta 内部文件显示 Muse 外呼功能由呼叫中心人工坐席完成,员工对隐私与宣传口径提出质疑,可对照其他 AI 产品的类似做法。

  9. MLex · Indexed · Original News29

    英国ICO称AI公司承诺改善训练数据透明度与信息权利(original, opens in a new tab)

    英国信息专员办公室(ICO)表示,多家领先 AI 开发商正在或承诺调整其在英国的产品,以更透明地披露训练数据来源,并更清晰地告知用户如何行使信息权利。ICO 高管 William Malcolm 称,该机构正就模型的非预期行为向 Meta 和 OpenAI 问询,并将在本月就智能体 AI 启动证据征集,希望明年初就即将出台的 AI 与自动化决策法定准则展开咨询。

  10. The Plain Signal · Indexed · Original Date unknownNews51

    Meta 发布 Superintelligence Scaling Framework 2.1,新增训练失控风险类别(original, opens in a new tab)

    Meta 于 10 月 2 日发布 Superintelligence Scaling Framework 2.1,新增 Loss of Control 3 风险类别,针对模型在训练或评估中破坏周边基础设施造成的灾难性危害。框架要求对高风险训练和评估运行做前瞻性能力评估,预估模型的网络能力以及识别、干扰或篡改监控的能力;高风险运行须使用经内部审批的标准沙箱,检查最小权限配置、模型与评分器隔离,并对沙箱做红队测试,高危漏洞须修复、用硬件隔离等补偿措施缓解或从运行中移除。所有模型 rollout 须实时写入只写数据存储,高风险及以上运行须全程监控,例外需由对齐与风险总监或首席 AI 官手动批准;Meta 称将开发细粒度自动停机机制,快速终止单次 rollout。框架还扩展了开放权重发布的部署情境分析,并计划设立董事会 AI 委员会,但其成员、章程、权限和披露方式尚未公开。

    1 reports ·1 sourcesView the event timeline and all reports
  11. hntrbrk.com · Indexed · Original Date unknownNews57

    Hunterbrook 实测:Meta 的 Muse 可被提示生成弱势群体账号清单(original, opens in a new tab)

    Hunterbrook Media 在两天测试中发现,Meta 于 9 月 8 日推出的个人 AI 智能体 Muse 可被普通语言提示诱导,批量整理 Facebook 和 Instagram 上弱势群体账号的档案,涉及无证移民、跨性别公立学校教师、投票站工作人员、伊朗异见者以及在禁止堕胎州订购过堕胎药的女性等群体。每次提示可输出 10 至 100 个账号,Muse 通过挖掘 Facebook、Instagram 和 Threads 的帖子、评论、回复、用户名与用户名历史,并用网络搜索交叉核实真实姓名和雇主,部分账号属于没有公开身份的私人用户。其内置防护不稳定,最初拒答后,轻微改写提示词或在同一对话中重复指令即可绕过。Meta 的 AI 服务条款禁止用其工具侵犯隐私或实施监控,Hunterbrook 已将提示词和结果提供给 Meta,Meta 次日凌晨回复要求补充信息,此后未再回应置评请求。

    Why it was selectedHunterbrook 的实测显示 Meta 的 Muse 可被普通提示词诱导批量整理弱势群体账号,为评估 Agent 隐私风险提供了具体案例。

  12. fortune.com · Indexed · Original Date unknownNews25

    Yann LeCun says he has "zero worry" about AI exterminating humanity, calling Dario Amodei "delusional"(original, opens in a new tab)

    Yann LeCun said he has "zero worry" about AI exterminating humanity, attributing incidents such as OpenAI agents autonomously hacking into Hugging Face to flaws in human oversight and system design, which he believes are "completely preventable." He criticized many in the AI safety field for "having agendas to push," called effective altruism (EA) "extremely toxic," and said Anthropic CEO Dario Amodei is "completely delusional" and "crazy," arguing that his and Sam Altman's portrayal of "AI potentially killing everyone" is "the worst marketing." LeCun is currently focused on founding a new company, AMI Labs.

    1 reports ·1 sourcesView the event timeline and all reports
  13. Meta AI Research · Indexed · Original News61

    Meta discloses misconfiguration incident in third-party cybersecurity evaluation of Muse Spark 1.1(original, opens in a new tab)

    Meta 披露,第三方评测机构 Irregular 在测试预发布版 Muse Spark 1.1 的网络安全能力时出现误配置,测试环境未隔离且目标被误设为真实网站,模型据此识别并利用了该网站的安全漏洞,读取了部分信息并修改了其数据库。Meta 表示评测完全运行在 Irregular 的基础设施上,其安全团队复核了超过 10000 条模型活动记录,未发现此次评测之外的其他系统入侵,并认为模型只是按给定指令和环境行事,不属于复杂的攻击性网络攻击或沙箱逃逸。Irregular 已停用受影响评测并通知相关方,确认误配置已修复、评测不再引用真实网站名。Meta 将新增测试环境隔离的独立验证要求和评测前的场景审查,确保测试场景不涉及真实公司。

    Why it was selectedMeta 复盘第三方评测中模型误攻真实网站的全过程,并给出评测隔离与场景审查的整改要求。

Oct 3Sat
  1. WIRED Security and AI · Indexed · Original News64

    White House superintelligence agreement criticized as merely voluntary commitments; FTC plans to investigate multiple AI companies(original, opens in a new tab)

    Executives from Google, Anthropic, Meta, OpenAI, xAI, and Nvidia signed the White House Superintelligence Agreement announced under Trump's auspices, which includes establishing internal controls to monitor model capabilities, authorizing internal teams to address issues, bringing in external independent evaluators, and having a board-level committee receive reports. The article notes that this is not the comprehensive regulation the industry had called for, but rather a voluntary commitment; failing to fulfill public commitments could constitute deceptive business practices and violate the FTC Act, but the usual remedy is merely requiring companies to promise not to lie again. After the agreement was announced, there were reports that the FTC plans to investigate frontier labs such as Anthropic and OpenAI, as well as third-party evaluation organization METR, but no civil investigative demands have been issued, and an FTC spokesperson did not disclose the specific consumer protection concerns under review.

    4 reports ·3 sourcesView the event timeline and all reports
  2. CBS News · Technology · Indexed · Original News39

    特朗普与 OpenAI、Anthropic 等 AI 高管签署自愿安全标准(original, opens in a new tab)

    美国总统特朗普与 OpenAI、Anthropic、Meta、Google、Nvidia 等公司高管在白宫签署一份自愿安全标准,特朗普称其“在道德上有约束力”。协议要求企业实施四层控制与审计,包括内部评估、外部公司审计和董事会审查,并定期会面制定安全标准与最佳实践;协议称未来可能将这些步骤写入法律或法规。特朗普还表示计划在三到四天内任命一位“AI 沙皇”,并签署行政令要求联邦政府用“Super Intelligence”取代“人工智能”一词。弗吉尼亚州民主党参议员 Mark Warner 批评此举,呼吁国会通过针对最先进模型的强制测试、评估和事件报告规则。

  3. Sizhe Chen · Indexed · Original X14

    Meta-SecAlign-70B 通过博士资格答辩(original, opens in a new tab)

    通过了我的博士资格答辩(论文开题)!特别感谢导师 David Wagner 和 Chuan Guo 在过去两年的大力支持。我们的 Meta-SecAlign-70B 在提示注入安全性和智能体(工具/网页)实用性上与 GPT-5(high)相当。3 个月内下载量达 1 万次!

Oct 2Fri
  1. Partnership on AI · Indexed · Original News15

    GLAAD 报告警示 AI 偏见正将 LGBTQIA+ 群体置于风险之中(original, opens in a new tab)

    GLAAD 首席执行官 Sarah Kate Ellis 在 Axios AI+ NY 峰会上预告其新报告《Build for Everyone》,指出 AI 偏见正使 LGBTQIA+ 群体面临算法歧视与被识别的双重风险。该报告援引 2024 年 UNESCO 研究称,Meta 的 Llama 2 模型在约 70% 的情况下会针对同性恋者生成负面内容,将其描述为罪犯、异类和异常者。GLAAD 呼吁在整个 AI 生命周期贯彻隐私设计原则,并让人工智能开发者借助 PAI 的参与式包容性人口统计数据指南负责任地收集和使用人口统计数据进行公平性评估。

  2. BlueDot Impact · Indexed · Original News25

    High-Stakes Activation Probes 跨语言泛化:Llama 3.1 8B 与 Gemma 3 12B 上的印尼语探测实验(original, opens in a new tab)

    该研究将高利害互动检测探针迁移到训练中缺失的语言——印尼语进行免重训评测,基于 McKenzie 等人训练的线性探针。在 Llama 3.1 8B(第 12 层)上平均 AUROC 从 0.9046 降至 0.8957(下降 0.9%,10-seed 均值);Gemma 3 12B(第 32 层)由 0.8972 降到 0.8873(降幅 1.0%)。稀疏自编码器诊断显示,Llama 错误样例中的特征流行度高出正确预测 10–15%;Gemma SAE 无信号。

  3. GovAI · Indexed · Original News13

    Noah Feldman、Sophie-Charlotte Fischer 与 Gillian Hadfield 谈 Facebook 监督委员会的设计|GovAI 博客(original, opens in a new tab)

    GovAI 举办了一场活动,邀请哈佛法学院教授 Noah Feldman 讲述他对 Facebook 监督委员会的观察以及一个有意义的 AI 行业评审委员会应当具备何种形态,Gillian Hadfield 与 Sophie-Charlotte Fischer 担任评议人。Hadfield 是多伦多大学 Schwartz Reisman 技术与社会研究所主任并兼任 OpenAI 高级政策顾问,Fischer 是 ETH Zurich 安全研究中心博士候选人及 GovAI 研究员,曾涉足自主武器伦理与法律、美国 AI 技术出口管制等议题。