Company Achievements & Recognition - Polyhedra team expresses gratitude for the support received throughout 2025 [1] - The year 2025 marked breakthroughs in verifiable AI and zkML, transitioning research into practical products like Polyhedra i-D [1] - Polyhedra acknowledges the builders and partners within its ecosystem for transforming foundations into real-world systems [1] Product Launch & Incentives - OpenCash System now features live private transactions [2] - A reward program offering 1 million ZKJ is available on OpenCash System during the holiday season [2] Future Outlook - Polyhedra expresses excitement for future advancements and developments [2] - The company envisions a future where verifiability is a key aspect [2]

zkML Benefits for Agentic Systems - zkML provides a proof layer for agentic systems, enhancing transparency and accountability [1] - Confirms the specific model version used for each decision [1] - Proves policy compliance without revealing internal logic [1] - Generates evidence of how multi-step actions were executed [1] - Enables auditors to validate behavior without accessing raw data [1] Enhanced Accountability and Verification - Transforms autonomous execution into a verifiable and reconstructable process [1] - Allows for independent verification and accountability of autonomous systems [1]

Industry Trend - Gartner predicts over 40% of agentic AI projects will be canceled by the end of 2027 [1] Technology & Control - The core issue is the lack of reliable verification methods for autonomous agents' actions post-deployment [1] - Enterprises lack control over autonomous agents [1]

AI Model Transparency and Auditability - Confirms reasoning steps followed required constraints [1] - Allows auditors to review the chain of logic without revealing proprietary model weights [1] - Turns AI-generated search results into something grounded — not improvised [1] - Confirms the model version responsible for each statement [2] - Confirms which evidence contributed to each claim, without exposing underlying files [2]

AI安全与风险 - AI生成内容可能存在捏造信息和虚假陈述的风险 [1] - 虚假信息在被发现后持续存在，问题解决存在滞后性 [1] 技术应用与挑战 - Google的AI搜索摘要出现问题，导致针对美国电气承包商的诉讼被捏造 [1] - 捏造信息甚至被错误地归因于州检察长 [1] 时间线与事件 - 公司在2024年发现了捏造的声明 [1] - 针对虚假信息的诉讼于2025年3月提起 [1] - 虚假输出持续到11月 [1]

zkML Importance - zkML 对于 AI agent 获得系统可信的身份至关重要 [1] AI Agent Security Features - 将 AI agent 身份绑定到加密凭证，而非 UI 会话 [2] - 证明哪个 agent 和模型实例执行了每个动作 [2] - 证明授权、支出限制和策略在执行时得到执行 [2] - 生成可验证的证据，而不暴露用户数据或内部逻辑 [2]

AI Agent Commerce - McKinsey 报告显示，AI agents 正在执行实际商业活动 [1] - OpenAI 的 Operator 可以完成预订和购买 [1] - Stripe 正在 ChatGPT 内部启用 agentic checkout（代理结账）[1] - Google 的 AP2 旨在让 agents 通过签名授权执行支付 [1] - AI agents 不仅仅是辅助，而是开始进行交易 [1]

zkML Overview - zkML directly addresses the mismatch between model execution and verification [1] - zkML enables proving which model executed a task, independent of hardware [1] - Verification becomes portable, even when infrastructure is not [1] Key Capabilities - zkML allows proving policy compliance without exposing model internals [2] - zkML facilitates proving how inputs shaped outputs without leaking data [2] - zkML ensures proving behavior consistency across decentralized deployments [2]

Model Performance - DeepSeek 发布了 V3.2 模型，在推理、数学和编码方面表现出色 [1] - 新模型显著降低了计算需求，使得前沿智能不再局限于超大规模基础设施 [1] Technological Advancement - DeepSeek 的 V3.2 模型实现了前沿级别的性能 [1]

zkML的核心价值 - zkML 为自主工作流执行提供可被检查、审计和证明的缺失层 [1] - zkML 能够证明工作流每个阶段执行的模型 [2] - zkML 能够证明所需的策略和约束已得到执行 [2] - zkML 能够在不暴露私有数据的情况下证明哪些输入影响了结果 [2] - zkML 使审计员能够验证多小时的代理追踪，而无需泄露专有逻辑 [2]