Blog | TikTok for Developers

TikTok for Developers

Our blog

Get insights into TikTok's technology, engineering updates, and more.

Articles

Beyond Hallucinations: Uncovering Linguistic Bias in Vision-Language ModelsAI-generated descriptions often get stuck in linguistic loops—repeating labels, flooding attributes, propagating roots. We detect these bias patterns using attribution before they skew creative outputs.

When Tools Become Prompts: Why Multi-Turn MCP Systems Break Traditional Security AssumptionsHow multi-turn MCP systems enable direct PII propagation and cross-server PII reconstruction, revealing security failures that do not appear in single-turn tool evaluations.

TikTok TechJam 2025 Highlights: Building with Joy, Coding for ChangeTikTok TechJam brings university students together to push boundaries and showcase creativity. This year's edition drew 2,000+ applications and 308 submissions, with 12 finalist teams tackling real-world challenges.

European Capture the Flag on Privacy-Preserving Database SystemsOblivious and TikTok announce PET ARENA, a CTF competition where participants design redteaming exercises against protections like differential privacy. Top prize: €4000. Open to students, researchers, and engineers based in Europe.

TikTok x IMDA Dev Day: Building with Singapore's Tech CommunityWe teamed up with Singapore's tech community at Developer Discovery Day to explore AI and cloud innovation, share TikTok's developer tools, and support hands-on learning and collaboration.

Introducing Knit: Zero-Overhead Dependency InjectionAs the TikTok app continued to grow, we began to encounter challenges that prompted us to rethink our approach to dependency injection.

TikTok saves up to 50% on costs with ~50 lines of code with Prompt CachingLearn how TikTok reduced inference costs by up to 50% using a lightweight prompt caching strategy implemented in just ~50 lines of code.

I ❤️ PETs Webinar: Exploring the Frontiers of AI, Identity and Privacy Through Open InnovationOn‑demand recap of the I ❤️ PETs half‑day webinar exploring privacy‑enhancing technologies for ethical AI, identity and trust.

PrivacyGo Joins IMDA PET Sandbox for Privacy-Preserving Ad MeasurementTikTok's open source PrivacyGo joins Singapore's IMDA PET Sandbox, enabling regulator-reviewed, privacy-preserving analytics with PETs like MPC and Differential Privacy for secure, compliant data collaboration.

LF AI & Data Day Singapore 2025: "Agent for SWE" RecapRecap of LF AI & Data Day Singapore 2025 where the open source AI community explored the "Agent for SWE" theme, covering code optimization, AI infrastructure insights.

ManaTEE: Enabling Verifiable AI Transparency via Confidential ComputingManaTEE securely evaluates AI models inside TEEs, preserving confidentiality while providing cryptographically verifiable results. Prove fairness and accuracy without exposing your model, ensuring trust between providers, partners, and regulators.

Transparency by Design: Why Open Source AI Matters for PrivacyAn overview of how open source AI supports algorithmic transparency, auditing, and regulatory compliance, highlighting its role in enabling oversight and aligning with emerging legal frameworks.

Showing 1 to 12 of 76 results