Launching Sora responsibly

Sora 2 与 Sora 应用将前沿的视频生成技术与一种全新的协作创作方式结合在一起，我们从一开始就把安全性内置其中。我们的做法以具体的保护措施为基础：

• 区分 AI 内容。所有由 Sora 生成的视频都包含可见和不可见的溯源信号。上线时，所有输出都带有可见水印。所有 Sora 视频还嵌入 C2PA 元数据（行业标准签名），我们内部还维护反向图像和音频搜索工具，能够高精度地将视频溯源回 Sora，构建在 ChatGPT 图像生成和 Sora 1 的成功系统之上。
• 基于同意的肖像使用。我们的目标是让你在整个流程中掌控自己的肖像（形象）。我们设置了护栏，旨在确保你的音频和图像肖像在获得你同意的情况下被使用，通过“客串”（cameo）功能实现。只有你可以决定谁能使用你的客串，并且你可以随时撤销访问权限。我们也采取措施阻止对公众人物的描绘（当然使用客串功能的除外）。包含你客串的视频——包括其他用户创建的草稿——始终对你可见，便于你查看并删除（如有需要还可举报）任何包含你客串的视频。我们还对任何包含客串的视频施加额外的安全护栏，你甚至可以为客串设定偏好设置，例如要求其始终戴一顶毡帽（fedora）。
• 针对青少年的保护措施。Sora 对未成年用户有更严格的保护，包括对成熟内容的限制。信息流设计适合青少年，青少年个人资料不会被推荐给成年人，成年人也不能主动向青少年发起消息。ChatGPT 中新增的家长控制允许家长管理青少年是否可以发送和接收私信，并可在 Sora 应用中为其选择非个性化信息流。默认情况下，青少年在 Sora 中也有连续滚动的次数/时长限制。
• 过滤有害内容。Sora 使用多层防护，在保留创意空间的同时维护信息流安全。在创作阶段，护栏旨在在内容生成前阻止不安全内容——包括性内容、恐怖主义宣传和自我伤害鼓吹——通过对提示和输出在多帧视频和音频转录中进行检查。我们进行了红队测试以探索新风险，并且鉴于 Sora 更高的逼真度以及动作与音频的加入，我们已相较于图像生成收紧了相关政策。除了生成环节之外，自动化系统会根据我们的全球使用政策（Global Usage Policies，https://openai.com/policies/usage-policies/）扫描所有信息流内容并过滤不安全或不适龄的材料。这些系统会随着我们对新风险的学习不断更新，并由重点针对高影响危害的人工作审查补充。
• 音频相关的安全措施。向 Sora 添加音频提高了安全要求，尽管完美的防护难以实现，我们仍在这一领域持续投入。Sora 会自动扫描生成语音的转录文本以发现潜在的政策违规，并阻止试图生成模仿在世艺术家或现有作品的音乐。我们的系统旨在检测并阻止此类提示，并会受理认为 Sora 输出侵权的创作者的下架请求。
• 用户控制与救济。你可以自行决定何时以及如何分享你的视频，并可随时删除已发布的内容。视频只有在你选择时才会被分享到信息流。每个视频、个人资料和评论都可以被举报滥用，一旦违反政策会有明确的处理渠道。你还可以随时屏蔽账号，屏蔽后对方将无法看到你的个人资料或通过私信联系你。

Sora 2 and the Sora app combine cutting-edge video generation with a new way to create together, and we’ve made sure safety is built in from the very start. Our approach is anchored in concrete protections:

• Distinguishing AI content.  Every video generated with Sora includes both visible and invisible provenance signals. At launch, all outputs carry a visible watermark. All Sora videos also embed C2PA metadata—an industry-standard signature—and we maintain internal reverse-image and audio search tools that can trace videos back to Sora with high accuracy, building on successful systems from ChatGPT image generation and Sora 1.
• Consent-based likeness. Our goal is to place you in control of your likeness end-to-end with Sora. We have guardrails intended to ensure that your audio and image likeness are used with your consent, via cameos. Only you decide who can use your cameo, and you can revoke access at any time. We also take measures to block depictions of public figures (except those using the cameos feature, of course). Videos that include your cameo—including drafts created by other users—are always visible to you. This lets you easily review and delete (and, if needed, report) any videos featuring your cameo. We also apply extra safety guardrails to any video with a cameo, and you can even set preferences for how your cameo behaves—for example, requesting that it always wears a fedora.
• Safeguards for teens. Sora includes stronger protections for younger users, including limitations on mature output. The feed is designed to be appropriate for teens, teen profiles are not recommended to adults, and adults cannot initiate messages with teens. New parental controls in ChatGPT let parents manage whether teens can send and receive DMs, as well as select a non-personalized feed in the Sora app. And by default, teens also have limits on how much they can continuously scroll in Sora.
• Filtering harmful content. Sora uses layered defenses to keep the feed safe while leaving room for creativity. At creation, guardrails seek to block unsafe content before it’s made—including sexual material, terrorist propaganda, and self-harm promotion—by checking both prompts and outputs across multiple video frames and audio transcripts. We’ve red teamed to explore novel risks, and we’ve tightened policies relative to image generation given Sora’s greater realism and the addition of motion and audio. Beyond generation, automated systems scan all feed content against our Global Usage Policies⁠ and filter out unsafe or age-inappropriate material. These systems are continuously updated as we learn about new risks and are complemented by human review focused on the highest-impact harms.
• Audio safeguards. Adding audio to Sora raises the bar for safety, and while perfect protections are difficult, we continue to invest seriously in this area. Sora automatically scans transcripts of generated speech for potential policy violations, and also blocks attempts to generate music that imitates living artists or existing works. Our systems are designed to detect and stop such prompts, and we honor takedown requests from creators who believe a Sora output infringes on their work.
• User control and recourse. You choose when and how to share your videos, and you can remove your published content at any time. Videos will only be shared to the feed when you choose to do so. Every video, profile, and comment can be reported for abuse, with clear recourse when policies are violated. You can also choose to block accounts at any time, which will prevent others from seeing your profile or contacting you via direct message.

Generated by RSStT. The copyright belongs to the original author.

Source