ByteDance sets up first-level AI Data & Safety department alongside Seed and Flow
On August 13, ByteDance announced a first-level AI Data & Safety department, integrating previously scattered data teams. The new unit sits alongside Seed and Flow, reports suggest founder Zhang Yiming opposes distilling external frontier models and the department will support the in-training 10-trillion-parameter model.
On August 13, ByteDance formally announced the establishment of a first-level AI Data & Safety department, sitting alongside Seed and Flow, consolidating data teams previously scattered across business lines. The new unit will provide cross-modal data support for an in-training 10-trillion-parameter super-large model.
Founder Zhang Yiming had previously told a Seed all-hands meeting that ByteDance does not rely on distilling external frontier models, and is devoting roughly half his time to Seed. CEO Liang Rubo acknowledged the gap with overseas top-tier models and called for accelerated catch-up. ByteDance's self-developed AI line has been advancing on two fronts: Seedance 2.5 and the consumer-facing Doubao (345 million monthly actives). The new department makes data a top-level asset in the org chart for the first time, closing the loop with Seed (fundamental research) and Flow (application layer).
From an industry perspective, ByteDance's reorganization reflects the trade-offs self-developed routes must make among data compliance, cross-modal alignment, and compute efficiency. If the 10-trillion-parameter figure holds, it would surpass current open-source leaders Kimi K3 (2.8 trillion) and DeepSeek V4-Pro (1.6 trillion), approaching the parameter count of Anthropic's Mythos family.