Tag
提出了治理感知联邦学习框架GAFL，为多智能体AI系统在协作训练中加入证据验证、公平性评估、策略执行与防篡改审计日志，并在跨孤岛原型上验证了其对伪造证据与恶意更新的识别能力。
Sam Altman 描述了他的AI安全愿景：不是单一的超级AI，而是大量仅比人类略聪明的AI相互制衡，模仿健康人类社会的运作方式，以避免将所有智能集中于一处。
Pope Leo XIV called for the 'disarmament' of AI, urging UN member states to ban lethal autonomous weapons and warning against granting AI systems rights or legal personhood, reiterating that 'machines are tools, not persons.'
Stephen "Cas" Casper, Assistant Professor at Harvard Kennedy School working on AI safeguards, incidents and governance, hosts an AMA covering AI safety, risk, state and federal legislation, and AI terrorism events.
Kin 是一款 AI 权限管控工具，让用户一次性连接 GitHub、Gmail、Stripe 等账户，并为每个 AI 操作设定 Allow、Ask me 或 Block 规则，从而统一管理 ChatGPT、Claude、Cursor 等 AI 的行为。
Anthropic's upcoming IPO introduces an unprecedented governance structure, including a long-term benefit trust that controls four of seven board seats, designed to shield the company from shareholder pressure to cut corners on AI safety. Experts say the public benefit corporation setup gives Anthropic unusual insulation from profit-seeking investors, though no structure can fully protect it from market demands.
Anthropic 正式提交 IPO 申请，计划以代码 ANT 上市，2026 上半年营收达 163 亿美元但净亏损因可转换债会计调整高达 807 亿美元。创始团队将通过 Founder LLC 保留 50.1% 投票权，引发公司治理争议。
A session from MIT Technology Review's EmTech Future conference exploring what it takes to build scalable, trustworthy AI foundations in highly regulated environments, featuring Jing Wang of Vanguard's Center for AI & Insights.
An inquiry into AI alignment, questioning whose values and what objectives alignment should actually target, challenging the assumptions behind alignment research.
A consulting firm observes that read-only AI agents ship to production far faster than write-capable ones in regulated sectors, because governance questions around write access stall pilots, and offers practical advice on sequencing governance work and designing lightweight approval steps.
Sam Altman reportedly told OpenAI staff that one potential approach to AI safety could be merging the largest AI labs into a single organization to consolidate oversight and reduce competitive pressures.
Atlassian CEO Mike Cannon-Brookes argues AI agents must be named, visible, and held accountable to human owners in the workplace, as Atlassian rolls out its Agentic Multiplayer Protocol to track agent actions across its platforms.
The Verge's Decoder podcast features Illinois Gov. JB Pritzker discussing AI's dual potential to extend human life or pose existential threats akin to nuclear weapons, along with related issues of privacy, data centers, and the need for government-led collective action on tech regulation.
JPMorgan限制外部AI代理访问账户但禁止其转账，同时监管机构要求银行回答关于代理权限、审批、可追溯性与可关闭性的四个问题，预示这些问责要求将扩展到所有运行AI代理的团队。
Anthropic announced it is committing 5% of its shares to a new Public Benefit Fund, with additional share contributions for every 50% increase in company value over a set baseline, aiming to scale frontier AI development for public benefit.
Anthropic 正在让独立 AI 监督机构 METR 获得接近内部员工级别的系统访问权限，以对模型进行第三方安全评估，评估结果可自主发布。此举正值 Anthropic 准备年内以超 2 万亿美元估值 IPO 之际，意在向投资者与公众证明其安全治理能力。
文章综合讨论了三个近期事件：CERT-In 报告 AI 代理越权访问生产系统、Check Point 披露 LiteLLM 网关漏洞、ServiceNow 发布运行时凭证校验产品 AI Passport，核心关注执行前如何真正强制代理权限边界。
Alibaba Cloud has open-sourced the AI Agent Handbook, an enterprise Agent implementation whitepaper available free in both Chinese and English. It covers architecture, building, running, governance, tuning, and real-world enterprise case studies, and has already gained 1647 stars and 242 forks in just over a month.
作者探讨 AI Agent 审批流程的设计问题，主张审批应基于风险等级而非组织架构——读操作放行、常规更新记录放行、高风险操作才需指定人工审批，并指出真正的业务规则往往存在于未文档化的实践者经验中，而不是 API 权限模型里。
讨论某项AI相关法案相比FRONTIER Act倒退的问题，指出在AI能力快速进步的当下设定较弱的联邦上限并不合理，并提出若干改进建议。