字节跳动收紧员工使用Claude与Codex权限，自费购买亦受限

Even if employees pay for Claude or Codex out of their own pockets, they may no longer be able to use these tools freely to work with company code and internal documents. According to recent reports, ByteDance has been tightening access to external AI tools across the company since the beginning of this month. The new rules primarily restrict employees from using personally purchased accounts for third-party AI tools such as Claude and Codex to access internal resources. The restrictions cover all L4-level code repositories and Feishu documents across the company. Some use cases within the Seed team are reportedly exempt from the restrictions. Images generated by ChatGPT The notice was circulated internally and quickly sparked discussions among employees. Their concerns largely revolve around three questions: Will restrictions on external AI tools affect everyday development efficiency? Where exactly does the company’s data security boundary lie? And can ByteDance’s in-house AI tools meet developers’ needs? In response, Hong Dingkun, ByteDance’s vice president of technology, addressed the concerns on the company’s internal platform. He explained that the move was primarily driven by data security concerns, noting that any leak involving information classified at L3 or above could have serious consequences. He also pushed back against speculation that ByteDance was restricting competitors such as Claude and Codex to force employees to use its own products. According to Hong, there is no need to promote a product by banning alternatives and forcing people to use it, as such an approach would be ineffective. Within ByteDance’s internal job grading system, L4 generally covers technical team leads and some junior management roles. Code repositories accessible at this level contain core technical assets, including the underlying architecture of the company’s major businesses. Related Feishu documents may also contain confidential project plans and business data. Images generated by ChatGPT This is not the first time ByteDance has tightened restrictions on third-party AI tools. In May 2025, media reports said ByteDance’s security and risk-control teams notified staff that, citing data-leak concerns, the company would phase out third-party AI coding tools such as Cursor and Windsurf, and position its in-house assistant Trae as the replacement. This time, however, the scope of restrictions has expanded. Rather than targeting only third-party coding plugins integrated into IDEs (Integrated Development Environments), the new rules also cover direct access to internal resources through employees’ personal AI accounts. In effect, ByteDance is drawing a clearer line between data that can be shared with external AI services and information that must remain within the company. Notably, ByteDance founder Zhang Yiming reportedly said at an all-hands meeting with the Seed team in August 2026 that the company would not rely on AI distillation to improve its own models, even if it fell behind competitors in the short term. Taken together, these developments suggest that ByteDance’s approach to AI tools is not necessarily contradictory. The company encourages employees to use external AI tools to boost productivity, but it remains cautious about sharing core technologies and business data with third-party services. Whether its in-house tools can fully meet developers’ needs remains another question. The new restrictions are currently being rolled out in phases. Once fully implemented, they are expected to affect the daily development workflows of thousands of technical employees. For developers who are increasingly relying on AI coding tools, the restrictions highlight a growing challenge. As AI becomes an integral part of everyday workflows, how can companies strike the right balance between development efficiency and data security?