AIP security and privacy（AIP 安全与隐私）¶

Palantir is committed to protecting the privacy and security of customer data. The protection and responsible processing of customer information are integral to our operations and built as first principles into our products, including AIP.

While generative AI models, including LLMs, present an opportunity to improve and accelerate business processes and decision-making, these technologies can raise issues around privacy and security, bias and discrimination, and the role of human judgment. Palantir takes these concerns seriously; this page contains a selection of frequently asked questions about AIP security and privacy. For more information, see FAQs: Security and Privacy of Palantir's AIP leveraging third-party-hosted LLMs ↗ in the Palantir trust portal ↗.

Is AIP protected by the same security measures as data in Foundry?¶

Yes. AIP incorporates all of Palantir's advanced security measures for the protection of sensitive data in compliance with industry regulations. AIP provides robust access controls, encryption, and auditing capabilities to maintain data integrity and transparency. Moreover, built-in governance tools help organizations maintain accountability and historical lineage in AI operations.

Where are AIP third-party-hosted models provided from?¶

When integrating with third-party-hosted models, AIP is designed to take advantage of regional endpoints wherever available and possible. Doing so helps minimize latency and is currently provided in regions like the US, UK and EU for some models. The specific geographic region is subject to the technical restrictions and stipulations that exist across the third-party-hosted AI model services available via AIP, which may change over time. Learn more about AIP geographic restrictions.

Does the service result in storage of any customer data by third-party-hosted model service providers?¶

No. When Palantir AIP leverages third-party-hosted model services, no customer data contained in prompts or completions is retained by the applicable third party.

Prior to making new models available in AIP, Palantir secures technical and contractual guarantees from third-party-hosted model service providers, ensuring consistent application of this policy.

Is customer data being used to retrain models?¶

No. When AIP accesses third-party-hosted model services, no customer data is used to retrain such models. Palantir secures strict technical and contractual guarantees from third-party-hosted model service providers to ensure that no customer data submitted in prompts or contained in completions is used for model training.

Completely separate from third-party-hosted model services: if customers do desire retraining of their own private AI model deployments, AIP can be used to facilitate such private, governed retraining, with full-spectrum governance tools for teams to audit, interrogate, and monitor model performance.

Do third-party-hosted model service providers have access to the data in AIP’s prompts?¶

No. No personnel of a third-party-hosted model service provider has access to prompts or completions, given the strict technical guarantees that Palantir ensures when establishing access for any given third-party-hosted model service. Third-party-hosted model service providers also do not store or retain customer prompts or completions. All data transmitted to the underlying services are immediately discarded after prompt completion.

Consult the relevant documentation for Azure ↗, AWS ↗, and Google Cloud ↗ for more information on associated privacy considerations with each provider.

How secure is AIP's leveraging of third-party-hosted model services?¶

Palantir AIP services are built atop secure infrastructure from cloud providers (inclusive of AWS, Azure, Google Cloud). Providers of third-party-hosted model services that have been made available through AIP, unless stated explicitly otherwise in your agreement with Palantir for AIP, have received ISO 27017, SOC (1, 2, 3), CSA STAR and/or other certifications. To learn more about Palantir’s security posture, visit Palantir’s trust portal ↗.

What contractual commitments has Palantir made around the processing of customer data through AI models and beyond?¶

Palantir signs data protection and substantially similar agreements (e.g., business associate agreements) before starting to process any personal data on customers’ behalf. These contractual commitments generally apply to all services provided by Palantir, including AIP.

How does Palantir enable responsible use of AI?¶

Palantir's Privacy and Civil Liberties Team has provided extensive guidance on developing, building, and deploying AI enabling technologies ↗. Palantir is committed to the following principles of AI ethics:

Focus on the fully integrated system, not just its component tools.
Acknowledge technology limits.
Don’t solve problems that shouldn’t be solved.
Adhere to methodological best practices for sound data science.
Keep AI responsible, accountable, and oriented towards humans.
Promote multi-stakeholder engagement.
Ensure technical, governance, and cultural awareness in data and technology applications.

Palantir's principles of security, privacy, and responsible use are the foundation of our product development and deployment.

中文翻译¶

AIP 安全与隐私¶

Palantir 致力于保护客户数据的隐私与安全。客户信息的保护与负责任处理是我们运营的核心，并作为首要原则融入包括 AIP 在内的所有产品中。

尽管生成式 AI 模型（包括 LLM）为改进和加速业务流程与决策提供了机遇，但这些技术也可能引发隐私与安全、偏见与歧视以及人类判断作用等问题。Palantir 严肃对待这些问题；本页面收录了关于 AIP 安全与隐私的常见问题精选。更多信息，请参阅 Palantir 信任门户 ↗ 中的常见问题解答：Palantir AIP 利用第三方托管 LLM 的安全与隐私 ↗。

AIP 是否受到与 Foundry 中数据相同的安全措施保护？¶

是的。 AIP 整合了 Palantir 所有先进的安全措施，以保护敏感数据并符合行业法规。AIP 提供强大的访问控制、加密和审计能力，确保数据的完整性和透明度。此外，内置的治理工具帮助组织在 AI 操作中保持问责性和历史溯源。

AIP 的第三方托管模型来自何处？¶

在集成第三方托管模型时，AIP 设计为尽可能利用区域端点。这样做有助于最小化延迟，目前在美国、英国和欧盟等地区为某些模型提供此服务。具体地理区域受 AIP 可用的第三方托管 AI 模型服务存在的技术限制和规定约束，这些限制和规定可能随时间变化。了解更多关于 AIP 地理限制的信息。

该服务是否会导致第三方托管模型服务提供商存储任何客户数据？¶

不会。 当 Palantir AIP 利用第三方托管模型服务时，提示或补全中包含的任何客户数据均不会被相关第三方保留。

在 AIP 中提供新模型之前，Palantir 会从第三方托管模型服务提供商处获得技术和合同保证，确保该政策的一致执行。

客户数据是否用于重新训练模型？¶

不会。 当 AIP 访问第三方托管模型服务时，没有任何客户数据被用于重新训练此类模型。Palantir 从第三方托管模型服务提供商处获得严格的技术和合同保证，确保提交在提示中或包含在补全中的任何客户数据均不用于模型训练。

与第三方托管模型服务完全独立：如果客户确实希望重新训练自己的私有 AI 模型部署，AIP 可用于促进此类私有的、受管控的重新训练，并提供全谱系治理工具供团队审计、查询和监控模型性能。

第三方托管模型服务提供商能否访问 AIP 提示中的数据？¶

不能。 鉴于 Palantir 在建立任何第三方托管模型服务访问权限时确保的严格技术保证，第三方托管模型服务提供商的任何人员均无法访问提示或补全。第三方托管模型服务提供商也不会存储或保留客户的提示或补全。传输到底层服务的所有数据在提示完成后立即丢弃。

请查阅 Azure ↗、AWS ↗ 和 Google Cloud ↗ 的相关文档，了解每个提供商相关的隐私注意事项。

AIP 利用第三方托管模型服务的安全性如何？¶

Palantir AIP 服务构建在云提供商（包括 AWS、Azure、Google Cloud）的安全基础设施之上。通过 AIP 提供的第三方托管模型服务提供商，除非您与 Palantir 就 AIP 达成的协议中另有明确说明，否则均已获得 ISO 27017、SOC（1、2、3）、CSA STAR 和/或其他认证。要了解更多关于 Palantir 安全态势的信息，请访问 Palantir 的信任门户 ↗。

Palantir 在通过 AI 模型及其他方式处理客户数据方面做出了哪些合同承诺？¶

Palantir 在开始代表客户处理任何个人数据之前，会签署数据保护及实质类似的协议（例如业务伙伴协议）。这些合同承诺通常适用于 Palantir 提供的所有服务，包括 AIP。

Palantir 如何实现 AI 的负责任使用？¶

Palantir 的隐私与公民自由团队已提供关于开发、构建和部署 AI 赋能技术的广泛指导 ↗。Palantir 致力于以下 AI 伦理原则：

关注完全集成的系统，而不仅仅是其组件工具。
承认技术局限性。
不解决不应解决的问题。
遵循方法论最佳实践，确保可靠的数据科学。
保持 AI 的负责任、可问责且以人为本。
促进多方利益相关者参与。
确保数据和技术应用中的技术、治理和文化意识。

Palantir 的安全、隐私和负责任使用原则是我们产品开发和部署的基础。