Encypher Readies C2PA Text Standard to Settle $5B AI Copyright Battles

Building Trust: Cryptographic Content Authenticity for the AI Era With Encypher — Ai4

Ai4

With probabilistic AI detectors failing at a 26% accuracy rate, embedding court-admissible cryptographic signatures into raw text provides the only deterministic solution for multi-billion-dollar copyright disputes.

AI-assisted summary of Building Trust: Cryptographic Content Authenticity for the AI Era With Encypher — original source. Check the original for full context.

Source published: 2025-08-24

Detailed notes

## Background Published on YouTube by Ai4 on August 24, 2025, this presentation features Erik Svilich, Founder and CEO of Encypher. According to the page description, Svilich explains how embedding invisible, tamper-evident cryptographic signatures directly into text under the emerging C2PA industry standard provides court-admissible evidence, creates a technical safe harbor for AI developers, and unlocks downstream model analytics.

## TL;DR Erik Svilich argues that with over $5 billion in active copyright litigation and conventional AI text detectors languishing at an abysmal 26% accuracy rate, the industry must rely on deterministic cryptographic proof rather than probabilistic heuristics. He showcases Encypher’s implementation of the C2PA text provenance framework, which embeds invisible SHA-256 hashes and digital signatures directly into unstructured text strings that survive web copy-pasting. Beyond establishing mathematical proof of origin and tampering in court, the technology delivers vital telemetry for AI labs to track model versions, prompt contexts, and viral engagement in the wild. Svilich highlights his co-chair role on the C2PA Text Task Force targeting a Q4 2025 standard release, alongside active enterprise pilots across publishing, defense, and healthcare.

## Atmosphere & style The presentation carries the brisk, authoritative momentum of a high-stakes tech keynote addressing a multi-billion-dollar legal crossroads. Svilich projects grounded confidence and urgency, playfully breaking the rule to 'never do a live demo' and embracing moments to 'nerd out' over cryptographic hashing. Watching the talk feels like witnessing an insider pivot point where courtroom panic gives way to pragmatic engineering standards.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) Over $5 billion in damages is currently contested in AI copyright lawsuits, yet existing AI text detectors remain fundamentally unreliable with roughly 26% accuracy. - [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) Mounting global regulations, including the EU AI Act and California provenance preservation laws, require deterministic, court-admissible proof rather than probabilistic detection. - [01:57](https://www.youtube.com/watch?v=Q8846epzTkQ&t=117s) Encypher establishes a cryptographic standard by embedding invisible, C2PA-compliant digital signatures directly into raw text strings. - [01:57](https://www.youtube.com/watch?v=Q8846epzTkQ&t=117s) Embedded cryptographic metadata survives copying and pasting across open web platforms such as LinkedIn and Medium without requiring proprietary file wrappers. - [03:43](https://www.youtube.com/watch?v=Q8846epzTkQ&t=223s) Cryptographic provenance provides downstream observability for AI labs, tracing viral content back to specific model versions, configurations, and prompt contexts. - [03:43](https://www.youtube.com/watch?v=Q8846epzTkQ&t=223s) Embedded text metadata solves a key architectural challenge for autonomous AI agents making decisions and communicating purely through unstructured text. - [05:59](https://www.youtube.com/watch?v=Q8846epzTkQ&t=359s) SHA-256 hashing within the C2PA manifest delivers mathematically verifiable tamper detection, flagging alterations down to a single modified word. - [05:59](https://www.youtube.com/watch?v=Q8846epzTkQ&t=359s) The C2PA Text Task Force, co-chaired by Encypher and Adobe, is establishing the open industry standard for text provenance with an official release targeted for Q4 2025. - [07:47](https://www.youtube.com/watch?v=Q8846epzTkQ&t=467s) Enterprise pilots for cryptographic text authenticity are expanding beyond publishing into defense recordkeeping, supply chain tracking, and clinical trial consent verification.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) The Multi-Billion Dollar Dispute and the Failure of AI Detectors

Erik Svilich, founder and CEO of Encypher, opens by framing the massive financial rift currently splitting the technology and media landscapes: "just over $5 billion" in damages is actively contested in court between publishers and the AI industry. High-profile lawsuits illustrate the immense scale of this conflict, with *The New York Times* seeking upwards of $4 billion in damages and Getty Images pursuing $1.7 billion. Svilich points out that the industry’s current defense mechanisms are fundamentally broken. Existing AI detection tools hover around an abysmal 26% accuracy rate, with the remaining results plagued by false positives—a failure rate so severe that Svilich jokes an organization would literally "save more money and be more accurate" simply by flipping a coin.

The legal and regulatory environment is intensifying this pressure. Regulators worldwide are cracking down on AI transparency and copyright infringement, from the European Union's EU AI Act—which introduces crippling fines of up to €35 million or 7% of global annual turnover—to the U.S. Securities and Exchange Commission levying fines exceeding $100,000 against companies making deceptive AI claims. Furthermore, newly passed legislation in California explicitly bars major media corporations from stripping provenance metadata out of content uploaded to their platforms. Against this backdrop, Svilich argues that relying on machine learning or heuristic algorithms to detect whether text was written by another AI is entirely futile; these probabilistic detectors are biased, unreliable, and cannot provide admissible legal proof in court.

- Current AI litigation damages: over $5 billion overall, including upwards of $4 billion sought by *The New York Times* and $1.7 billion by Getty Images. - Conventional AI text detectors: approximately 26% accuracy, with the vast majority of non-matches being false positives. - Regulatory penalties: EU AI Act fines up to €35 million or 7% of worldwide turnover; SEC penalties exceeding $100,000; new California provenance preservation laws.

---

[01:57](https://www.youtube.com/watch?v=Q8846epzTkQ&t=117s) The "Technical Peace Treaty": C2PA Cryptographic Signatures in Raw Text

To resolve this impasse, Svilich proposes a "technical peace treaty between publishers and AI labs": a unified cryptographic infrastructure that bridges copyright compliance with commercial utility. Acknowledging the old conference speaker adage to "never do a live demo," he proceeds to do one anyway, showcasing Encypher's system running inside standardized industry formats. He demonstrates an article snippet attributed to Encypher Corporation using the metadata manifest structure developed by the Coalition for Content Provenance and Authenticity (C2PA).

```mermaid flowchart TD accTitle: Cryptographic Text Provenance Flow subgraph Origin["1. Source Generation"] A["Original Text Content"] --> B["SHA-256 Hashing"] B --> C["C2PA Manifest & Digital Signature"] end subgraph Distribution["2. Open Web Distribution"] C --> D["Embedded Invisible Metadata"] D --> E["Unstructured Text on LinkedIn / Medium"] end subgraph Verification["3. Verification & Insights"] E --> F["Encypher Decoder Engine"] F --> G["Proof of Origin & Tamper Status"] F --> H["Model Performance & Engagement Analytics"] end ```

The system packages content identity, claim generators, and assertions directly into the text object and cryptographically signs them to verify the author's authenticity beyond dispute. Svilich emphasizes that this signature survives even when content is copied and pasted as unstructured text onto platforms like LinkedIn or Medium. Because the metadata remains invisibly embedded inside the raw text string itself, anyone can paste an article into the Encypher decoder to verify its origin and chain of custody instantly. The software is already live in production environments alongside active design partners.

- Standardization body: Coalition for Content Provenance and Authenticity (C2PA). - Delivery mechanism: Cryptographically signed metadata embedded directly within raw, unstructured text strings without requiring proprietary file wrappers. - Live verification: Demonstrable by copy-pasting Svilich's recent LinkedIn publications directly into the Encypher decoder.

---

[03:43](https://www.youtube.com/watch?v=Q8846epzTkQ&t=223s) From Evidence to Business Intelligence: Model Observability in the Wild

Beyond serving as courtroom evidence, Svilich explains that provenance metadata unlocks vital business intelligence for AI developers. When an organization embeds cryptographic tracking into model outputs, it gains granular visibility into model performance across versions, whether comparing GPT-3 and GPT-3.5 or navigating the transition between GPT-4.5 and GPT-5. He connects this to a major architectural hurdle highlighted by Mastercard earlier at the conference: autonomous AI agents making financial decisions and communicating entirely in unstructured text, completely divorced from traditional file formats. Without embedded metadata, tracking origin and agent authorization across unstructured data streams is practically impossible.

Svilich illustrates the commercial value of this traceability through the lens of viral content. If an author writes an article that goes viral using an AI model, embedded cryptographic provenance allows the AI lab to trace that viral artifact all the way back to the specific model version, configuration settings, and prompt context that produced it. This fills a critical visibility blind spot for AI companies that invest heavily in research and development—averaging around $2.7 billion per frontier model version—yet currently operate with zero post-generation observability once users copy and paste text across the web. The same cryptographic signature that protects intellectual property simultaneously provides the telemetry required to optimize models for real-world user engagement.

- Average frontier AI research and development cost: approximately $2.7 billion per model generation. - Operational challenge: AI systems and autonomous agents operate on unstructured text lacking native container formats. - Dual-use utility: Cryptographic signatures simultaneously establish legal IP ownership and deliver downstream telemetry on real-world engagement and prompt configurations.

---

[05:59](https://www.youtube.com/watch?v=Q8846epzTkQ&t=359s) Mathematical Tamper Detection and the C2PA Text Standard Roadmap

Svilich dives into the underlying cryptography, pausing to "nerd out for a little bit" on Encypher's tamper-detection pipeline. The platform computes a cryptographic SHA-256 hash across the original text—such as the classic pangram "The quick brown fox jumps over the lazy dog"—generating an immutable digital digest that functions like a mathematical password. If even a single word is altered (for example, swapping "fox" for "cat"), re-running the modified text through the algorithm produces an entirely mismatched hash. This discrepancy instantly flags that a specific sentence, paragraph, or article has been altered.

Because this hash is embedded within the C2PA-compliant cryptographic manifest, the resulting proof of tampering and proof of origin is mathematically absolute and admissible in court, unlike heuristic AI detectors. Svilich reveals that he serves as co-chair of the C2PA Text Task Force alongside Leonard Rosenthol, Senior Principal Engineer at Adobe. The task force is formalizing the open, consensus-based standard for embedding cryptographic metadata directly into text, with the official specification scheduled for release in the fourth quarter of 2025. Svilich notes that this framework is designed to provide the ultimate technical resolution for protracted copyright battles, citing *The New York Times* spending over $11 million on legal fees alone in a single year during its battle against OpenAI.

- Leadership: Erik Svilich and Leonard Rosenthol (Adobe) co-chairing the C2PA Text Task Force. - Target milestone: Official C2PA text specification release set for Q4 2025. - Industry litigation costs: *The New York Times* incurred upwards of $11 million in legal fees in a single year fighting its OpenAI lawsuit.

---

[07:47](https://www.youtube.com/watch?v=Q8846epzTkQ&t=467s) Enterprise Adoption: Pilots Across Publishing, Defense, and Healthcare

Svilich concludes by outlining Encypher’s current enterprise footprint and operational roadmap. Contrasting Encypher’s open-source core and deterministic 100% verification rate against heuristic black-box models, he details ongoing pilots with major publishing houses and Fortune 500 corporations trying to de-risk multi-billion-dollar copyright liabilities.

Beyond media, the technology is expanding into high-stakes regulatory and defense domains. Svilich notes active discussions with former U.S. Department of Defense and CIA personnel in Estonia around secure government recordkeeping, regulatory compliance, and supply chain provenance for critical inventory tracking. In healthcare, the cryptographic architecture is being adapted to guarantee chain of custody for patient consent records and clinical trial documentation. Svilich emphasizes that early adopters are actively establishing market precedent for AI content provenance. He closes with an open invitation for partnerships, noting that after filling an agreement earlier that day, Encypher has only three remaining pilot slots open for media publishers and one remaining integration slot for frontier AI labs.

- Active enterprise pilots: Fortune 500 enterprises and major commercial publishers. - Specialized sector use cases: - Government and intelligence: Secure public records and compliance tracking alongside ex-DoD/CIA contacts in Estonia. - Logistics: Tamper-evident supply chain tracking for critical inventory. - Healthcare: Verifiable patient consent tracking and clinical trial data provenance. - Availability: Open pilot capacity limited to three media publishers and one AI lab partner.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) just over $5 billion — Erik Svilich

> [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) save more money and be more accurate — Erik Svilich

> [01:57](https://www.youtube.com/watch?v=Q8846epzTkQ&t=117s) technical peace treaty between publishers and AI labs — Erik Svilich

> [01:57](https://www.youtube.com/watch?v=Q8846epzTkQ&t=117s) never do a live demo — Erik Svilich

> [05:59](https://www.youtube.com/watch?v=Q8846epzTkQ&t=359s) nerd out for a little bit — Erik Svilich

> [05:59](https://www.youtube.com/watch?v=Q8846epzTkQ&t=359s) The quick brown fox jumps over the lazy dog — Erik Svilich

## Mentioned Erik Svilich, Encypher, The New York Times, Getty Images, OpenAI, Anthropic, Universal Music, Adobe, Meta, Google, Microsoft, BBC, EU AI Act, Securities and Exchange Commission (SEC), California, Coalition for Content Provenance and Authenticity (C2PA), LinkedIn, Medium, Mastercard, GPT-3, GPT-3.5, GPT-4.5, GPT-5, SHA-256, Leonard Rosenthol, C2PA Text Task Force, U.S. Department of Defense, CIA, Estonia, Fortune 500

## Video description Erik Svilich, Founder & CEO, Encypher

Abstracts The AI industry's multi-billion dollar lawsuits, from The New York Times vs. OpenAI to Universal Music vs. Anthropic, won't be settled by today's unreliable AI detectors. They will be decided by cryptographic proof.

I will demonstrate how we embed invisible, tamper-evident signatures directly into text to provide court-admissible evidence that transforms copyright disputes into enforceable licensing agreements. You will see the specific implementation that is becoming the C2PA industry standard backed by Adobe, Meta, Google, OpenAI, Microsoft, the BBC, and others.

Learn how this technology offers a technical safe harbor for AI developers, enables data analytics on your model's performance in the wild, and creates a monetization path for content creators, enabling a new economy for licensed content built on trust and transparency. This is your opportunity to see the infrastructure moving from courtrooms to mainstream workflows.

[Source](https://www.youtube.com/watch?v=Q8846epzTkQ)

中文简报

Encypher推进C2PA文本标准以化解50亿美元AI版权争端在概率性AI检测器准确率低至26%的情况下，在原始文本中嵌入法庭认可的加密签名，为数十亿美元的版权纠纷提供了唯一的确定性解决方案。 Erik Svilich认为，当前涉及的版权诉讼金额超过50亿美元，而传统AI文本检测器的准确率仅有惨淡的26%，业界必须依靠确定性的密码学证明，而非概率启发式方法。他展示了Encypher对C2PA文本溯源框架的实现方案，该方案将不可见的SHA-256哈希和数字签名直接嵌入非结构化文本字符串中，即便在网络上复制粘贴也能完整留存。除了在法庭上提供关于来源和篡改的数学证明外，该技术还为AI实验室提供了关键的遥测能力，以追踪实际运行中的模型版本、提示词上下文和病毒式传播参与度。Svilich还强调了其作为C2PA文本工作组联合主席的身份，该工作组计划于2025年第四季度发布标准，并已在出版、国防和医疗保健领域展开积极的企业试点。目前 AI 领域的版权诉讼索赔总额已超 50 亿美元，而传统 AI 文本检测器的准确率仅约 26%，随着欧盟 AI 法案及加利福尼亚州溯源法规落地，行业亟需具备法庭采信效力的确定性验证手段。 (0:04) Encypher 联合推进基于 C2PA 规范的密码学方案，将不可见的数字签名直接嵌入纯文本字符串中，使其在 LinkedIn 或 Medium 等开放平台被复制粘贴后依然能完整保留溯源信息。 (1:57) 文本元数据不仅能解决自主 AI Agent 在非结构化交互中的鉴权难题，还填补了 AI 实验室在内容生成后的可观测性盲区，支持将传播内容反向回溯至具体的模型版本和 Prompt 配置。 (3:43) 通过在 C2PA Manifest 中运用 SHA-256 哈希算法，该技术可对细微至单个词的篡改实现数学级检测；Encypher 与 Adobe 共同主持的 C2PA 文本工作组计划于 2025 年第四季度正式发布该行业开源标准。 (5:59)

Reading summary

Erik Svilich, founder and CEO of Encypher, argues that probabilistic AI detection is failing amidst billions in copyright litigation, necessitating an open cryptographic standard that embeds tamper-evident provenance directly into unstructured text. This framework is designed to provide legally defensible proof of origin while unlocking downstream operational telemetry for frontier AI developers. The Breakdown of Heuristic AI Detection (0:04) The intersection of media and artificial intelligence faces severe legal friction, with actively contested damages exceeding $5 billion across major lawsuits, such as The New York Times seeking over $4 billion and Getty Images pursuing $1.7 billion. Defending against these challenges has created major cost burdens, including The New York Times incurring upwards of $11 million in legal fees in a single year. Concurrently, regulatory authorities are increasing compliance requirements: the EU AI Act mandates penalties up to €35 million or 7% of global turnover, the SEC has penalized deceptive AI claims, and California has enacted legislation barring platforms from stripping content provenance metadata. Svilich asserts that conventional AI detectors cannot resolve these disputes because statistical heuristics achieve only around 26% accuracy, with the remaining results plagued by false positives. Because probabilistic tools cannot establish admissible legal proof in court, Svilich contends that relying on machine learning to detect AI-generated text is fundamentally ineffective. C2PA Standards and Cryptographic Text Hashing (1:57) As an alternative to statistical detection, Encypher utilizes the metadata manifest framework established by the Coalition for Content Provenance and Authenticity (C2PA). The system embeds cryptographically signed provenance data directly into raw, unstructured text strings without requiring proprietary file wrappers. This design allows invisible metadata—including author assertions and claim generators—to persist when content is copied and pasted across public web platforms such as LinkedIn or Medium, where it can be decoded to verify origin. To establish tamper detection, Encypher computes a cryptographic SHA-256 hash of the source text to generate an immutable digital digest. Altering even a single word invalidates the matching hash, providing mathematical proof of modification that Svilich notes is admissible in legal proceedings. Svilich co-chairs the C2PA Text Task Force alongside Adobe's Leonard Rosenthol, with the consensus-backed open standard scheduled for formal specification release in the fourth quarter of 2025. Downstream Model Telemetry and Agent Observability (3:43) Beyond legal evidence, cryptographic text provenance offers technical observability for AI labs. Svilich observes that frontier AI model development costs approximately $2.7 billion per generation, yet labs historically lose all visibility into output utilization once text is copied into the wild. Cryptographic signatures allow developers to trace real-world content back to specific model versions, configurations, and prompt contexts, delivering telemetry to evaluate real-world engagement. This provenance architecture also addresses governance issues raised regarding autonomous AI agents. As highlighted by Mastercard at the conference, autonomous agents increasingly communicate and make financial decisions entirely in unstructured text. Svilich notes that without native cryptographic metadata embedded inside these text streams, verifying authorization and tracking transaction origin across agent workflows remains virtually impossible. Enterprise Expansion Across Healthcare and Defense (7:47) Encypher is running early enterprise pilots with Fortune 500 companies and major publishers seeking to mitigate copyright liabilities. According to Svilich, early adopters are establishing precedent for deterministic, 100% verification rather than relying on black-box heuristic scanners. Outside commercial publishing, the underlying architecture is being adapted for regulatory and security applications. Encypher is engaging with former U.S. Department of Defense and CIA personnel in Estonia regarding government recordkeeping, regulatory compliance, and critical inventory supply chain tracking. In healthcare, the protocol is targeted at securing patient consent trails and clinical trial documentation. While Svilich positions cryptographic signatures as a definitive resolution to copyright and observability disputes, the framework's broad utility hinges on cross-industry adoption and the formalization of the C2PA text standard slated for late 2025.

精炼阅读摘要

Encypher 创始人兼首席执行官 Erik Svilich 指出，面对超过 50 亿美元的版权争议与严苛的全球监管，传统概率型 AI 检测工具因极低准确率已无法作为合规防线；行业亟需通过 C2PA 标准在非结构化文本中嵌入密码学数字签名，以建立具备法庭效力的数学级溯源与模型观测体系。巨额版权诉讼与传统 AI 检测的失效 (0:04) Svilich 指出，科技界与媒体出版行业目前正陷入耗资巨大的法律冲突，法庭上争议的赔偿金额已累计超过 50 亿美元。典型案件包括《纽约时报》针对侵权索赔超过 40 亿美元，且其单年诉讼支出就已突破 1100 万美元；盖蒂图片社（Getty Images）的诉讼索赔亦达到 17 亿美元。与此同时，全球监管正在快速收紧：欧盟《人工智能法案》（EU AI Act）针对违规行为设立了最高达 3500 万欧元或全球年营业额 7% 的巨额罚金；美国证券交易委员会（SEC）对虚假 AI 宣传开出了超过 10 万美元的罚单；加利福尼亚州的新立法更明确禁止主流媒体平台擅自剥离内容附带的溯源元数据。面对法律与合规压力，Svilich 强调现有防御机制已彻底失效。市面上的传统机器学习与启发式 AI 文本检测工具准确率仅在 26% 左右，其余判定绝大部分为假阳性误报。Svilich 戏称，企业依靠抛硬币来判断文本出处甚至比使用这些检测器更省钱、更准确。由于此类概率工具存在固有算法偏差，无法在法庭上作为可采信的合法证据，依赖它们根本无法抵御日益严峻的版权侵权指控。 C2PA 文本加密机制与防篡改标准 (1:57) 为打破出版界与 AI 实验室的对立僵局，Svilich 提出了以密码学基础设施为核心的“技术和平协议”。该方案采用内容来源和真实性联盟（C2PA）的元数据清单规范，利用 SHA-256 加密算法对原始文本进行数学哈希计算。以经典英文全字母句“The quick brown fox jumps over the lazy dog”为例，若其中任意字符被替换（例如将“fox”篡改为“cat”），算法重新计算生成的哈希值将发生完全改变。这种数学机制能瞬间锁定句子、段落或篇章级别的篡改行为，提供不同于黑盒概率模型的、在法律上绝对可信的防篡改证明。与以往依赖专有容器的多媒体凭证不同，Encypher 技术将 C2PA 加密签名直接隐形嵌入于原始、非结构化的纯文本字符串中。在现场演示中，Svilich 展示了即使文本被复制并粘贴至 LinkedIn 或 Medium 等开放平台，底层元数据依旧得以完整保留，用户通过解码器即可瞬时还原其真实出处与流转链条。Svilich 透露，他目前正与 Adobe 高级首席工程师 Leonard Rosenthol 共同担任 C2PA 文本工作组联合主席，推动该开源共识标准的建立，官方文本规范预计将于 2025 年第四季度正式发布。非结构化文本的商业智能与遥测 (3:43) Svilich 进一步阐述，加密溯源不仅是法律诉讼中的举证工具，更能为 AI 研发机构提供关键的商业智能。目前，行业训练前沿基础模型的单代平均成本高达约 27 亿美元，但模型一旦生成文本并被用户复制传播到开放网络后，开发机构便彻底失去了下游使用与传播的可见性。在输出文本中内嵌加密溯源机制后，AI 实验室可跨模型版本（如 GPT-3 与 GPT-3.5，或 GPT-4.5 与 GPT-5）追踪实际表现，填补这一高昂的研发观测盲区。这种追踪能力在商业层面极具价值。例如，某篇基于 AI 生成的文章在网络走红后，加密凭证能够让实验室逆向追溯该内容所对应的具体模型版本、超参数配置及提示词上下文。Svilich 还结合万事达卡（Mastercard）在会议中提及的痛点指出，未来执行金融决策并开展交互的自主 AI Agent 普遍通过非结构化纯文本进行通信，完全脱离了传统文件格式。若无文本内嵌元数据，系统将难以校验 Agent 的身份授权与指令链条。密码学文本签名由此同时承担了知识产权确权与下游生产数据遥测的双重职能。出版、医疗与国防领域的落地试点 (7:47) 在落地推进方面，Svilich 介绍称，Encypher 依托其确定性的 100% 验证率（当元数据清单存在时）及开源核心，已与多家财富 500 强企业和大型商业出版集团建立试点合作，协助企业规避潜在的高额侵权负债。部分早期设计合作伙伴已将该方案部署于生产环境，在真实业务流中验证内容溯源机制。除数字出版行业外，该加密架构正在向高合规门槛的政企和国防领域扩展。Svilich 透露，公司正与身处爱沙尼亚的前美国国防部（DoD）及中央情报局（CIA）背景人员展开合作沟通，计划将该技术引入敏感政府记录的安全存档、法规合规核查以及关键物资的供应链防伪追溯中。在医疗健康领域，该方案正用于确保患者知情同意书记录的不可伪造性以及临床试验数据的链路完整性。Svilich 表示，由于早期合作伙伴正在确立市场先例，当前对外部机构开放的试点席位已所剩无几，仅余三个媒体出版商名额和一个前沿 AI 实验室集成名额。 Svilich 认为，基于密码学的确定性验证终将替代黑盒概率检测工具，成为调和版权方与 AI 实验室矛盾的底层基石。不过，该体系能否真正实现全球规模的互信，仍取决于定于 2025 年第四季度发布的 C2PA 文本行业标准能否获得跨平台软件与主流大模型生态的广泛采纳与系统级支持。

Detailed notes — zh

## Background 本视频由 Encypher 创始人兼首席执行官 Erik Svilich 于 2025 年 8 月 24 日在 Ai4 平台发布。视频介绍展示了如何通过在文本中直接嵌入不可篡改的隐形密码学签名（C2PA 行业标准），替代不可靠的传统 AI 检测工具，化解数十亿美元的版权侵权诉讼，并为 AI 开发者建立合规技术安全港与模型分发分析能力。 ## TL;DR 面对传统 AI 检测器仅约 26% 的极低准确率与高昂的行业诉讼风险，Encypher 推出了基于密码学溯源的文本确权方案。该技术采用 SHA-256 与 C2PA 联盟标准，攻克了非结构化文本在跨平台流转时易丢失元数据的难题，实现 100% 确定性的防篡改与来源验证。此外，该加密签名还能转化为商业智能洞察，帮助大模型研发团队全链路追踪内容传播与模型性能，推动从版权死结走向可信许可经济。 ## Atmosphere & style 演讲整体节奏紧凑、直击痛点，兼具前沿技术的硬核底气与商业落地的现实感。Erik Svilich 语言风趣幽默且充满自信，现场敢于打破常规进行实时 Demo 演示，用严谨的数学逻辑彻底解构传统统计学检测工具的弊端。整场分享散发着一种开创行业新标准、重塑数字内容信任秩序的号召力，带给观众一种见证基础设施关键变革的临场紧迫感。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) 出版传媒业与 AI 行业面临超过 50 亿美元的版权侵权纠纷，而传统基于概率的 AI 文本检测器准确率仅约 26%，误报严重且缺乏法庭采信力。 - [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) 全球合规监管日益严苛，欧盟《人工智能法案》违规罚金最高达 3500 万欧元或全球营业额 7%，加州立法也明确禁止剥离内容版权溯源信息。 - [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) Encypher 提出“技术和平协议”，主张通过在生成源头嵌入不可篡改的加密签名，为出版商与 AI 厂商构建确定性的密码学信任基础设施。 - [02:21](https://www.youtube.com/watch?v=Q8846epzTkQ&t=141s) 系统通过 C2PA 标准清单结构封装生成声明与归属信息，攻克了非结构化文本缺乏文件容器、在流转中易丢失元数据的溯源痛点。 - [02:21](https://www.youtube.com/watch?v=Q8846epzTkQ&t=141s) 即便文本被复制粘贴至 LinkedIn、Medium 等第三方外部平台，解码器仍可完整提取元数据，为金融等领域 AI Agent 的授权与决策提供可信依据。 - [04:45](https://www.youtube.com/watch?v=Q8846epzTkQ&t=285s) 头部科技企业每代大模型平均研发投入高达 27 亿美元，但长期面临模型输出内容交付后在互联网流转传播难以追踪的致命盲区。 - [04:45](https://www.youtube.com/watch?v=Q8846epzTkQ&t=285s) 加密签名可实现从爆款内容全链路反查具体的底层模型版本、推理参数与提示词上下文，将版权凭证转化为指导模型研发优化的商业智能。 - [06:01](https://www.youtube.com/watch?v=Q8846epzTkQ&t=361s) 系统基于 SHA-256 密码学哈希比对，任何单字符的微小篡改都会导致哈希雪崩校验失败，从而提供具备数学确定性的司法级防篡改证据。 - [06:01](https://www.youtube.com/watch?v=Q8846epzTkQ&t=361s) Encypher 联合 Adobe 等巨头牵头推进 C2PA 文本工作组行业规范，获得 OpenAI、Google、微软、Meta 等联盟支持，官方规范将于 2025 年 Q4 正式推出。 - [07:49](https://www.youtube.com/watch?v=Q8846epzTkQ&t=469s) 与统计学黑盒检测方案相比，基于开源核心的 Encypher 在元数据存在时可提供 100% 确定性验证，目前已在多家媒体与《财富》500 强企业展开生产试点。 - [07:49](https://www.youtube.com/watch?v=Q8846epzTkQ&t=469s) 该密码学溯源技术正快速延伸至政府安全合规归档、高价值供应链库存管理以及医疗患者知情同意书与临床试验追踪等多元场景。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) AI 时代的数十亿美元诉讼危机与“技术和平协议” 在开场中，Encypher 的创始人兼首席执行官 Erik Svilich 抛出了一个令人震惊的数字：出版传媒行业与人工智能领域之间目前正面临着超过 50 亿美元的法律诉讼。其中，《纽约时报》（The New York Times）针对 OpenAI 提起的侵权索赔高达 40 亿美元以上，Getty Images 针对 AI 公司的诉讼索赔也达 17 亿美元。然而，目前市场上主流的 AI 文本检测工具准确率仅有约 26%，其余大部分情况下都会产生严重的误报（False Positives）。Erik 直言不讳地调侃道，面对这种现状，用户“还不如直接抛硬币，不仅更省钱，准确率可能还更高”。基于概率算法和机器学习黑盒去检测另一个伪装成人类写作的 AI 输出，既不可靠也充满偏差，根本无法在法庭上作为实质性证据。与此同时，全球监管压力正在急剧上升。欧盟的《人工智能法案》（EU AI Act）规定了最高 3500 万欧元或全球营业额 7%（以较高者为准）的巨额违规罚款；美国证券交易委员会（SEC）对虚假 AI 陈述的处罚也动辄超过 10 万美元；加利福尼亚州近期更通过了相关法规，严禁大型媒体公司从上传至其系统的内容中剥离版权溯源信息（Content Provenance）。面对技术与法律的双重死结，Erik 提出了一种基于密码学溯源的“技术和平协议”（Technical Peace Treaty）：通过在内容生成的源头嵌入不可篡改的加密签名，为出版商与 AI 实验室之间搭建一座通用信任基础设施，既满足合规性审查，又为商业智能分析奠定基础。 - **行业诉讼规模**：出版传媒业与 AI 行业纠纷索赔超 50 亿美元（《纽约时报》索赔逾 40 亿美元，Getty 索赔 17 亿美元）。 - **现有检测器痛点**：传统 AI 检测工具准确率仅约 26%，误报率极高，缺乏法庭采信力。 - **监管合规风险**：欧盟《人工智能法案》罚金高达 3500 万欧元或 7%；加州立法禁止剥离内容溯源信息。 --- ### [02:21](https://www.youtube.com/watch?v=Q8846epzTkQ&t=141s) 突破非结构化文本瓶颈：C2PA 文本加密签名与跨平台验证 Erik 随后带来了一段现场演示，并打趣道自己违背了导师“在大会演讲时千万别做实时 Demo”的忠告。他展示了一段由 AI 辅助生成的文本片段，并通过 C2PA（内容来源与真实性联盟，Coalition for Content Provenance and Authenticity）标准清单结构对其进行封装。在这一对象中，清晰记录了生成工具（Encypher Corporation Claim Generator）以及具体的生成行为与归属声明。随后，系统利用数字签名对其进行密码学层面的签署，从数学上不可辩驳地证明“作者确实是他本人”。这一突破的关键在于解决了“非结构化文本”（Unstructured Data）的溯源难题。以往图像或视频可以通过特定的文件格式容器（如 JPEG、MP4 的元数据块）携带溯源信息，但文本通常缺乏固有文件格式，极易在流转中丢失元数据。Encypher 的技术使得用户即便将文本复制粘贴到 Medium、LinkedIn 等第三方平台，只要将公开文本粘贴回解码器，依然能完整提取并验证嵌入的加密元数据。这种无缝跨平台溯源能力对于 AI Agent 尤为关键——正如万事达卡（Mastercard）在同日演讲中所探讨的课题，金融机构需要确切证明某个代表个人或企业进行决策并执行支付操作的智能体确实拥有合规授权与可信来源。 ```mermaid flowchart LR accTitle: C2PA 文本加密与跨平台验证流程 subgraph Origin["内容生成与签署"] A["原始内容 / AI 生成文本"] --> B["构建 C2PA 标准清单"] B --> C["数字签名与元数据隐形嵌入"] end subgraph Distribution["自由分发与流转"] C --> D["无格式文本复制粘贴"] D --> E["发布至 LinkedIn / Medium 等平台"] end subgraph Verification["溯源与验证"] E --> F["Encypher 解码器"] F --> G["100% 确定性来源与防篡改验证"] end ``` --- ### [04:45](https://www.youtube.com/watch?v=Q8846epzTkQ&t=285s) 证据转化为商业智能：大模型研发投资的性能回溯与优化嵌入文本的加密元数据不仅是法律维权证据，还能直接转化为高价值的业务洞察（Business Intelligence）。头部科技企业在每个大模型版本上的平均研发投入高达 27 亿美元，但各大厂商普遍面临一个致命盲区：模型交付给用户后，根本无法追踪用户生成的内容在互联网上的实际表现与传播路径。从 ChatGPT-4.5 市场反响平平，到 GPT-5 引发的广泛争议，AI 厂商亟需明确了解用户是如何复制、二次加工以及跨平台分发模型内容的。通过 Encypher 的加密签名，内容发布者与模型开发者能够获得前所未有的“性能洞察”（Performance Insights）。当某篇由特定模型（无论是版本 3、3.5 还是后续迭代）辅助生成的文章在社交网络引发病毒式传播时，开发者可以通过签名全链路回溯到源头，精准获知触发该爆款内容所使用的具体模型版本、推理参数配置以及完整的上下文环境。这一机制将版权保护与数据分析合二为一，使得同一套签名既能守护知识产权，又能反哺模型研发，指导团队优化用户互动体验。 - **大模型单版本研发成本**：主要科技巨头平均每代模型 R&D 支出约为 27 亿美元。 - **端到端传播追踪**：支持从爆款内容一键反查底层模型版本、推理参数与上下文提示词。 - **双重价值闭环**：密码学凭证兼具法庭级版权保护与产品级分发数据洞察功能。 --- ### [06:01](https://www.youtube.com/watch?v=Q8846epzTkQ&t=361s) SHA-256 防篡改验证与 C2PA 文本行业标准落地 Erik 深入讲解了系统的防篡改数学原理。以经典测试句 "The quick brown fox jumps over the lazy dog" 为例，系统会对其运行 SHA-256 密码学哈希算法，生成一段类似于不可逆密码的哈希指纹，并嵌入到 C2PA 加密对象中。一旦文本被任何人做出微小篡改——例如将其中的 "fox" 改为 "cat"——再次执行哈希计算时，输出的哈希值就会发生雪崩式变化，与原始签名不再匹配。这种机制使系统能够以数学确定性精确判定文章、段落甚至句子的篡改痕迹，提供在司法审判中坚不可摧的法庭证据。为推动该技术的全球标准化，Erik 以联合主席的身份领导了 C2PA 的文本工作组（Text Task Force），与 Adobe 的高级首席工程师 Leonard 共同牵头制定该规范。该联盟汇聚了 Adobe、Meta、Google、OpenAI、微软、BBC 等全球巨头。Erik 指出，《纽约时报》仅在去年一年就为与 OpenAI 的诉讼耗费了超过 1100 万美元的律师费，而 C2PA 文本标准正是彻底化解此类高昂版权死锁的行业级框架。该标准的正式官方规范将于 2025 年第四季度正式发布。 - **防篡改核心机制**：基于 SHA-256 哈希比对，任何单字符修改均会导致密码学校验失败。 - **联盟生态支持**：C2PA 及内容真实性倡议（CAI）由 Adobe、Google、微软、OpenAI、Meta、BBC 等联合支持。 - **标准时间表**：C2PA 文本嵌入官方开源规范定于 2025 年 Q4 正式推出；《纽约时报》单年诉讼律师费曾超 1100 万美元。 --- ### [07:49](https://www.youtube.com/watch?v=Q8846epzTkQ&t=469s) 开源核心与先发优势：跨行业试点与早期生态招募在对比市场方案时，Erik 强调了传统统计学检测模型与 Encypher 之间的本质差异：前者属于不可解释的黑盒猜测，准确率徘徊在 26% 左右；而 Encypher 是基于开源核心的生产级就绪方案，在元数据存在的情况下能够达到 100% 的绝对验证准确率。目前，该方案已在多家媒体机构与《财富》500 强企业中展开生产试点，直接攻坚 AI 领域的数十亿美元级痛点。除了传媒与大模型领域，该基础设施还迅速拓展至多个高价值应用场景： 1. **政府与合规安全**：与前美国国防部（DoD）及 CIA 驻爱沙尼亚专家合作，用于安全档案记录与合规性审查； 2. **供应链关键溯源**：对高价值及关键库存流转进行全生命周期溯源； 3. **医疗与生命科学**：实现患者知情同意书（Patient Consent）管理及临床试验数据的不可篡改追踪。在演讲尾声，Erik 宣布基础设施已全面就绪，现场公开招募首批试点生态伙伴——当天出版商试点已锁定 1 席，仅剩 3 个出版商席位，以及最后 1 个面向 AI 基础模型厂商的输出端集成席位。他呼吁行业先锋抓住这一确立市场先例的历史机遇，共同构建基于信任与透明度的新型数字内容经济。 ## Notable quotes > The AI industry's multi-billion dollar lawsuits, from The New York Times vs. OpenAI to Universal Music vs. Anthropic, won't be settled by today's unreliable AI detectors. They will be decided by cryptographic proof. — Erik Svilich > I will demonstrate how we embed invisible, tamper-evident signatures directly into text to provide court-admissible evidence that transforms copyright disputes into enforceable licensing agreements. — Erik Svilich > [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) 出版传媒行业与人工智能领域之间目前正面临着超过 50 亿美元的法律诉讼。 — Erik Svilich > [00:04](https://www.youtube.com/watch?v=Q8846epzTkQ&t=4s) 还不如直接抛硬币，不仅更省钱，准确率可能还更高。 — Erik Svilich > [02:21](https://www.youtube.com/watch?v=Q8846epzTkQ&t=141s) 在大会演讲时千万别做实时 Demo。 — Erik Svilich > [04:45](https://www.youtube.com/watch?v=Q8846epzTkQ&t=285s) 头部科技企业在每个大模型版本上的平均研发投入高达 27 亿美元，但各大厂商普遍面临一个致命盲区：模型交付给用户后，根本无法追踪用户生成的内容在互联网上的实际表现与传播路径。 — Erik Svilich > [06:01](https://www.youtube.com/watch?v=Q8846epzTkQ&t=361s) The quick brown fox jumps over the lazy dog — Erik Svilich > [06:01](https://www.youtube.com/watch?v=Q8846epzTkQ&t=361s) 《纽约时报》仅在去年一年就为与 OpenAI 的诉讼耗费了超过 1100 万美元的律师费，而 C2PA 文本标准正是彻底化解此类高昂版权死锁的行业级框架。 — Erik Svilich > [07:49](https://www.youtube.com/watch?v=Q8846epzTkQ&t=469s) 当天出版商试点已锁定 1 席，仅剩 3 个出版商席位，以及最后 1 个面向 AI 基础模型厂商的输出端集成席位。 — Erik Svilich ## Mentioned Erik Svilich, Encypher, The New York Times, OpenAI, Universal Music, Anthropic, Getty Images, EU AI Act, SEC, C2PA, Adobe, Meta, Google, Microsoft, BBC, Mastercard, Medium, LinkedIn, ChatGPT-4.5, GPT-5, SHA-256, Content Authenticity Initiative (CAI), Leonard, DoD, CIA ## Video description Erik Svilich, Founder & CEO, Encypher Abstracts The AI industry's multi-billion dollar lawsuits, from The New York Times vs. OpenAI to Universal Music vs. Anthropic, won't be settled by today's unreliable AI detectors. They will be decided by cryptographic proof. I will demonstrate how we embed invisible, tamper-evident signatures directly into text to provide court-admissible evidence that transforms copyright disputes into enforceable licensing agreements. You will see the specific implementation that is becoming the C2PA industry standard backed by Adobe, Meta, Google, OpenAI, Microsoft, the BBC, and others. Learn how this technology offers a technical safe harbor for AI developers, enables data analytics on your model's performance in the wild, and creates a monetization path for content creators, enabling a new economy for licensed content built on trust and transparency. This is your opportunity to see the infrastructure moving from courtrooms to mainstream workflows. [Source](https://www.youtube.com/watch?v=Q8846epzTkQ)

Source quotes

Erik Svilich: The AI industry's multi-billion dollar lawsuits, from The New York Times vs. OpenAI to Universal Music vs. Anthropic, won't be settled by today's unreliable AI detectors. They will be decided by cryptographic proof. Erik Svilich: I will demonstrate how we embed invisible, tamper-evident signatures directly into text to provide court-admissible evidence that transforms copyright disputes into enforceable licensing agreements. [0:04] Erik Svilich: 出版传媒行业与人工智能领域之间目前正面临着超过 50 亿美元的法律诉讼。 [0:04] Erik Svilich: 还不如直接抛硬币，不仅更省钱，准确率可能还更高。 [2:21] Erik Svilich: 在大会演讲时千万别做实时 Demo。 [4:45] Erik Svilich: 头部科技企业在每个大模型版本上的平均研发投入高达 27 亿美元，但各大厂商普遍面临一个致命盲区：模型交付给用户后，根本无法追踪用户生成的内容在互联网上的实际表现与传播路径。 [6:01] Erik Svilich: The quick brown fox jumps over the lazy dog [6:01] Erik Svilich: 《纽约时报》仅在去年一年就为与 OpenAI 的诉讼耗费了超过 1100 万美元的律师费，而 C2PA 文本标准正是彻底化解此类高昂版权死锁的行业级框架。 [7:49] Erik Svilich: 当天出版商试点已锁定 1 席，仅剩 3 个出版商席位，以及最后 1 个面向 AI 基础模型厂商的输出端集成席位。

Source quotes

[0:04] Erik Svilich: just over $5 billion [0:04] Erik Svilich: save more money and be more accurate [1:57] Erik Svilich: technical peace treaty between publishers and AI labs [1:57] Erik Svilich: never do a live demo [5:59] Erik Svilich: nerd out for a little bit [5:59] Erik Svilich: The quick brown fox jumps over the lazy dog

Original transcript

[0:04] Hello everyone. Eric Cilich, uh founder [0:07] and CEO of Incipher. [0:10] Um [0:12] $5 billion or just over $5 billion. Uh [0:16] this is the amount that's being sued [0:19] over between publishing and media [0:21] organizations and the AI industry with [0:23] New York Times seeking upwards of $4 [0:25] billion in damages. Getty $1.7 billion [0:28] in damages. And uh there's a real big [0:31] industry problem here between publishing [0:33] and media industry and AI companies. Um [0:38] currently there are [0:41] uh AI detection tools out there which [0:43] are about 26% accurate. And I made a [0:46] mistake on this slide. Uh [0:49] because the other percent is false [0:51] positives. So you would be better off [0:53] flipping a coin. it would save you more [0:55] money and be more accurate than current [0:58] AI detection tools out there. Uh the [1:01] solution that I'm going to walk you [1:02] through today and part of my standards [1:04] work with C2PA is a method of [1:08] cryptographic verification embedded at [1:10] the source of origin for content. [1:14] So there's a lot of regulation in this [1:17] uh space as well uh with the EUA act [1:20] with $ 35 million euro uh euro fines or [1:24] 7% whichever is greater and the SEC [1:26] currently finding companies upwards of [1:28] $100,000 [1:30] and also regulation in California that [1:34] just passed preventing large media [1:37] corporations from scraping content [1:39] providence out of content uploaded into [1:41] their systems. um detection tools, you [1:44] know, algorithms, ML or AI that's trying [1:47] to detect another AI output that's [1:50] trying to pretend that they're writing [1:52] human content are just unreliable and [1:54] biased. They're not proof. [1:57] So what I'm proposing is a technical [2:00] peace treaty between publishers and AI [2:02] labs where there's one infrastructure [2:05] that bridges the gap between publishing [2:08] organizations and AI labs that there's [2:11] one signature that not only enables [2:14] compliance but also business [2:16] intelligence which is very important. [2:18] I'll get into that in a second. So this [2:21] is an oversimplified demo. Uh one of my [2:23] good mentors who speaks at conferences [2:25] says never do a live demo. So that's [2:26] what I'm going to do here. Uh so for the [2:29] first part here is an example article by [2:32] me you know just a little snippet where [2:33] text has been generated and then we have [2:37] a C2PA which is the coalition for [2:40] content providence and authenticity [2:41] organization uh standard uh manifest [2:45] structure where essentially I have the [2:47] company which is incipher corporation a [2:50] claim gen generator and an action saying [2:53] that this piece of content has been [2:55] generated by me essentially [2:58] So we make this object that is part of [3:00] the industry standard and uh we actually [3:03] sign it using digital signatures which [3:05] is a cryptographic way to verify I am [3:08] who I say I am and then you can copy and [3:12] paste this content onto online platforms [3:15] let's say Medium or LinkedIn you can [3:17] copy paste my most recent LinkedIn [3:19] article into my decoder and see this [3:20] metadata embedded into it and then you [3:23] can have a cryptographic verification of [3:26] where this content is coming from which [3:29] is very important for trust. Um it's [3:32] currently uh I have some design [3:34] partnerships. It is running in some [3:35] production infrastructure right now and [3:38] uh like I said you can copy and paste my [3:39] article on LinkedIn and see it working [3:41] right now. [3:43] So this evidence becomes business [3:46] intelligence. So this we saw right here [3:48] there's metadata embedded into this [3:50] article. Why is this important? Well you [3:53] can run performance insights. Let's say [3:55] part of this article was generated by a [3:57] specific model and let's say there's [4:00] version three of the model and version [4:02] 3.5 of the model you know chat GPT is [4:04] all the way up to five. Uh so you can [4:07] not only check for proof of origin if [4:10] you are an AI company or let's say a [4:12] social media platform or if you are a [4:15] payment processor trying to verify that [4:17] this is your agent. There was a great [4:19] great talk by Mastercard earlier today. [4:21] This is a problem they're trying to [4:22] solve to verify this agent on behalf of [4:25] you or your organization is going out [4:27] and making a decision. Well, how do you [4:29] have proof of origin? These guys are [4:31] using unstructured data. They don't use [4:33] file formats, right? And all these [4:36] agents and uh artificial intelligence [4:39] systems use text which inherently [4:42] doesn't always have the file format. And [4:44] let's say my article goes viral and I [4:48] made it using a c in part with a certain [4:51] model. I can trace that viral content [4:53] all the way back to its source. The [4:55] exact model that was used, the settings, [4:58] all the content uh or all the context [5:00] that was made to make that viral output. [5:04] So essentially enabling enabling [5:06] performance insights on how my models [5:08] are working for me are performing. Uh [5:10] this is also really important because on [5:12] average uh the R&D spend by major [5:15] companies for each model version is [5:17] about $2.7 billion and they do not have [5:21] performance insights on how their [5:22] content is being used across the web. We [5:24] saw chat GPD's 4.5 model being a flop [5:27] and five you know there's arguments [5:29] whether it's a flop or been successful. [5:31] You may have your own opinion, but part [5:34] of the problem there is they don't have [5:35] performance insights as to how users are [5:38] copying and pasting that content and [5:40] making content that other users actually [5:42] enjoy interacting with, which is core to [5:45] their business. [5:47] So that same signature that not only [5:49] proves and protects your IP and your [5:52] content, also tells you how to optimize [5:55] for better engagement. [5:57] Pretty cool. [5:59] Also, I have the tamper detection built [6:01] in. Uh, I'm going to nerd on out on you [6:04] for a little bit, but see, we have some [6:06] example original text here. Quick brown [6:09] fox jumps over the lazy dog. So, part of [6:11] what we do is we run a cryptographic [6:13] hash algorithm, which is a Shaw 256 [6:16] hash, which basically converts this [6:19] group of text into almost like a [6:21] password. And if you were to modify that [6:25] in any way and say the quick brown cat [6:28] instead of the fox jumps over the lady [6:30] uh lazy dog, if you run it through the [6:32] same algorithm, that password will no [6:34] longer match. So you know that that [6:36] sentence or that paragraph or that [6:38] article was modified in some way and [6:40] this gets embedded into our [6:42] cryptographic object that we saw before [6:44] that is compliant with the industry [6:45] standard. Um, so essentially you have [6:50] proof of tampering and proof of origin [6:52] that is mathematically verifiable, holds [6:55] up in court, and something that [6:57] detectors can't currently be used as [6:59] evidence, which is part of why these [7:00] lawsuits are still ongoing. Uh so as [7:03] part of my effort uh I've joined C2PA [7:06] and the content authentic authenticity [7:08] initiative backed by Adobe, Google, [7:10] Microsoft, BBC open AAI and uh I'm the [7:13] co-chair of the text task force for C2PA [7:17] uh which is deploying in production [7:19] infrastructure today and for instance [7:22] when the New York Times and OpenAI [7:23] finally settled their over a year-long [7:26] lawsuit that they spent over $11 million [7:28] on legal fees alone New York Times last [7:30] year on this lawsuit. This will be the [7:33] framework that that solves this. U I'm [7:36] co-chairing the task force with the [7:38] senior principal engineer at Adobe uh [7:40] Leonard. So in quarter 4 2025 later this [7:43] year, expect that to be uh the official [7:45] spec. Uh so theory meets reality. So [7:50] like I said, quarter 4 of 2025, that's [7:53] when uh the text standard for embedding [7:54] metadata into text will be released. Uh [7:56] it's an open standard and consensus [7:58] based. you know, we vote to ratify the [8:00] standard and uh cryptographic proof [8:02] baseline. So, you have actual proof of [8:04] origin. Also, competitors uh 26% [8:08] accuracy. Uh there statistical models. [8:11] It's a black box and there's no proof, [8:13] right? You're guessing. Like I said, you [8:15] would save more money with a coin flip. [8:18] Uh and then there's incipher. We're [8:20] production ready 100% accuracy when the [8:22] metadata is there. We're an open source [8:25] core and we're available today. Uh also [8:29] uh we are currently doing media pilots [8:32] and Fortune 500 pilots uh that are [8:34] underway which is solving a [8:35] multi-billion dollar problem. That's a [8:37] daily conversation with these [8:38] organizations. There's also some [8:40] government use cases. I'm talking to [8:42] some XDOD CIA people in Estonia about uh [8:45] for secure records and compliance also [8:47] supply chain providence and tracking for [8:49] critical inventory and for healthcare [8:52] for patient consent also clinical trials [8:54] and that sort of providence. Um these f [8:57] first movers are helping solve a [8:59] multi-billion dollar problem, probably [9:00] the number one problem that AI has uh [9:03] now and setting the market precedent [9:05] across the industry. [9:07] So the infrastructure it's ready right [9:10] now. Uh I'm currently have three more [9:13] pilots for publishers. I already got one [9:15] today and I have one more slot for AI [9:17] companies who want to embed data into [9:19] their model outputs. And uh I invite you [9:21] to lead the standard and and start [9:24] owning this new market that's emerging. [9:26] I encourage you to connect with me on [9:28] LinkedIn, book a demo, and uh happy to [9:30] answer any questions you guys have after [9:32] this. Thank you so much. Yes.