> Markdown version of [/videos/1036-tiktok-s-privacy-innovation?t=1081](https://www.wearedevelopers.com/videos/1036-tiktok-s-privacy-innovation?t=1081). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# TikTok's Privacy Innovation

How do you train accurate models without exposing raw data? TikTok solved this. Their open-source clean room uses synthetic data and hardware isolation to guarantee zero leakage.

- **Speakers:** [Mingshen Sun](https://www.wearedevelopers.com/@mingshen-sun)
- **Event:** World Congress 2024
- **Published:** August 20, 2024
- **Duration:** 20:46
- **URL:** https://www.wearedevelopers.com/videos/1036-tiktok-s-privacy-innovation

## Summary

Navigating data collaboration presents a significant challenge when attempting to balance usability, security, accurate analytics, and privacy policy enforcement. Traditional solutions often force unavoidable trade-offs; SQL-based policy clean rooms offer high accuracy but struggle with robust privacy enforcement, while purely mathematical differential privacy often requires sacrificing data fidelity. Furthermore, hardware-level protections like Trusted Execution Environments (TEEs), despite being highly secure, usually lack the interactive user experience required for dynamic data analysis and experimental coding. To overcome these infrastructural gaps, the TikTok Privacy Innovation Lab developed an open-source framework designed to merge accessible interfaces with rigorous protection protocols.

To resolve the inherent tension between interactive research and strict data security, the solution implements a novel two-stage data clean room platform. The initial "programming stage" provisions data consumers with a familiar Jupyter Notebook interface equipped with differentially private synthetic data or completely randomized public sets. This operational separation enables researchers to securely explore patterns and develop machine learning models—such as XGBoost algorithms—without ever exposing sensitive, raw information. Once the exploratory code is verified, the job seamlessly transitions to the "execution stage," where it automatically runs against the full dataset inside a computationally isolated, hardware-attestable TEE. By bifurcating the workflow, organizations successfully minimize privacy risks during the experimental process while fully guaranteeing accuracy and zero-data-leakage during the final output generation.

Engineered for scalable deployment across modern cloud providers, this infrastructure acts as an ideal foundation for building Trusted Research Environments (TREs). Such secure workspaces allow external contributors to measure advertisements, trace economic impacts, or conduct privacy-preserving machine learning without conflicts of interest. Crucially, the platform issues cryptographic attestation reports—downloadable JSON Web Tokens (JWT)—giving external researchers independent verification that their logic was executed securely within a protected confidential space. Looking forward, the initiative plans to augment these capabilities by automating synthetic data generation, embracing intricate multi-way provider collaborations, and integrating modern GPUs to accelerate complex confidential computing workloads.

**Keywords:** data collaboration frameworks, privacy enhancing technologies, data clean rooms, trusted execution environments, differential privacy, synthetic data generation, confidential computing, interactive programming interfaces, attestation reports, trusted research environments, privacy preserving machine learning, hardware isolation, cloud infrastructure security, multi-way data collaboration

## Chapters

1. **Data collaboration workflows and multi-party architectures** (00:02) — Enabling multiple data providers and consumers to interact via controlled platform policies.
1. **Security and privacy goals for data collaboration platforms** (02:28) — Balancing data confidentiality and targeted access controls with interactive usability and accuracy.
1. **Evaluating existing privacy enhancing technology limitations** (04:26) — Standard approaches like SQL clean rooms, differential privacy, and raw hardware isolation fail to balance usability with strict privacy guarantees.
1. **Designing a two-stage confidential data clean room** (07:07) — Splitting workloads into a mock-data programming stage and a secure-hardware execution stage preserves both analytic flexibility and data security.
1. **Guaranteeing integrity through hardware trusted execution environments** (09:49) — Integrating hardware-level isolation allows programmatic access to accurate datasets without exposing the raw information to individual researchers.
1. **Cloud infrastructure deployment and industry use cases** (11:31) — Leveraging cloud-provided confidential compute instances supports secure research environments, precise ad measurement, and privacy-preserving machine learning.
1. **Configuring synthetic data for safe interactive programming** (13:50) — Processing public datasets through high-quality differential privacy algorithms generates safe proxies for early-stage workload development.
1. **Executing remote data exploration and model training** (15:00) — Researchers analyze data distributions and schedule predictive machine learning jobs securely onto isolated backend hardware environments.
1. **Validating hardware execution using secure attestation tokens** (17:01) — Downloading and verifying cryptographically signed hardware reports ensures analytic workloads run exclusively inside a verified confidential space.
1. **Advancing confidential computing with open source multi-way collaboration** (18:01) — Supporting multi-party architectures alongside automated synthetic data provisioning and upcoming isolated GPU computing capabilities broadens trusted execution adoption.

## Related Moments

- [Blending trusted hardware with future security strategies](https://www.wearedevelopers.com/videos/574-this-machine-ends-data-breaches) (from "This Machine Ends Data Breaches")
- [Navigating data privacy boundaries and adversarial model reliability](https://www.wearedevelopers.com/videos/260-getting-started-with-machine-learning) (from "Getting Started with Machine Learning")
- [Embedding data security and applied ethics into developer education](https://www.wearedevelopers.com/videos/1107-the-future-of-developer-experience-with-genai-driving-engineering-excellence) (from "The Future of Developer Experience with GenAI: Driving Engineering Excellence")
- [Securing data in use with confidential cloud computing](https://www.wearedevelopers.com/videos/100108-building-sovereign-ai-lessons-from-deploying-secure-rag-systems-using-confidential-computing) (from "Building Sovereign AI: Lessons from Deploying Secure RAG Systems using Confidential Computing")
- [Demonstrating transparency and privacy protection in machine learning](https://www.wearedevelopers.com/videos/509-a-walkthrough-on-responsible-ai-frameworks-and-case-studies) (from "A walkthrough on Responsible AI Frameworks and Case Studies")
- [Addressing audience concerns on licensing and privacy](https://www.wearedevelopers.com/videos/522-how-we-will-build-the-software-of-tomorrow) (from "How we will build the software of tomorrow")

## Related Articles

- [Stephan Gillich - Bringing AI Everywhere](https://www.wearedevelopers.com/magazine/489-stephan-gillich-bringing-ai-everywhere)
- [Panel Discussion: Responsible AI in Practice - Real-World Examples and Challenges](https://www.wearedevelopers.com/magazine/488-panel-discussion-responsible-ai-in-practice-real-world-examples-and-challenges)
- [Dev Digest 134 - Where pixels sing?](https://www.wearedevelopers.com/magazine/477-dev-digest-134-where-pixels-sing)
- [Dev Digest 138 - Are you secure about this?](https://www.wearedevelopers.com/magazine/486-dev-digest-138-are-you-secure-about-this)

## Related Jobs

- [TikTok Shop - Data Scientist (EMEA)](https://www.wearedevelopers.com/jobs/ext/1912878-tiktok-shop-data-scientist-emea) at **TikTok**
- [Staff Developer Advocate, GitHub Security Lab](https://www.wearedevelopers.com/jobs/ext/1921051-staff-developer-advocate-github-security-lab) at **GitHub**
- [Security Architect - AI](https://www.wearedevelopers.com/jobs/ext/1581899-security-architect-ai) at **ZEISS Group**
- [Principal Software Engineer, Enterprise AI Platform](https://www.wearedevelopers.com/jobs/ext/1467292-principal-software-engineer-enterprise-ai-platform) at **GitHub**
- [Data Scientist](https://www.wearedevelopers.com/jobs/ext/1351648-data-scientist) at **Almedia**
- [Lead Software Engineer - Data Engineering](https://www.wearedevelopers.com/jobs/ext/2000968-lead-software-engineer-data-engineering) at **Dynatrace**