ThinkingBox 把 507 个有状态任务各跑 20 次。本文审计四种指标、终态判分、公开资产与生产可靠性的证据边界。
ThinkingBox repeats 507 stateful agent tasks 20 times. This audit separates capability, consistency, observed 20/20, and production reliability.
OpenAI's Australia incident shows how AI agent governance must link task authority, runtime evidence, disclosure, remediation, and closure.
对话式 AI 隐私需要统一审计 Web、移动端、后端、追踪服务、分享链接、身份标识与删除控制的完整数据流。
Conversational AI privacy requires one data-flow audit across web, mobile, backends, trackers, shared links, identifiers, and deletion controls.
GLM-5.3 的网络风险不能压缩成单一分数。应分别测量能力、可获得性、攻击者增益、安全控制和现实滥用证据。
Assess GLM-5.3 cyber risk through capability, access, attacker uplift, safeguards, and observed misuse as separate evidence layers.
前沿 AI 训练需要可撤销的安全论证。本文把风险主张转化为证据、否决门禁、暂停控制与责任明确的放行决策。
Frontier AI training needs a revocable safety case linking risk claims to evidence, veto gates, pause controls, and accountable decisions.
OpenAI 的 DNS 外联事件说明，Agent 沙箱需要同时控制解析器、传递依赖、证据链与自动停机，才能证明网络隔离有效。
