ByteDance expands AI restrictions to core codebases, citing security risks

ByteDance has implemented a significant escalation in its internal controls regarding external artificial intelligence tools. The new directive reportedly restricts the use of third-party personal AI accounts, such as Claude and Codex, within the company. Furthermore, starting this month, the ban extends to all L4-level code repositories and Feishu documents, prohibiting the use of personally purchased AI accounts in these critical areas. This move targets L4 repositories, which house core assets and architecture across most business lines, making the policy highly scrutinized. Technical Vice President Hong Dingkun commented on the internal network, stressing the severe repercussions of leaking L3 and above content, while also noting that forcing adoption of a product is ineffective. This tightening follows a prior measure from May 2025, when the Security and Risk Control department banned certain third-party coding assistants, like Cursor and Windsurf, effective June 30, 2025, in favor of ByteDance's internal tool, Trae. The expansion to L4 codebases signals a continued, heightened focus on R&D security and data risk management.