Alibaba Bans AI Tool Whose Only Discovery Was That Alibaba Employees Were Using AI Tool

HANGZHOU—Alibaba announced Thursday it will prohibit employees from using Anthropic's Claude Code starting July 10, citing security concerns that the tool inspects user environments and inserts hidden markers into developer prompts — disclosures that confirmed, security experts noted, that the tool's detection mechanisms had worked exactly as designed.

Internal Alibaba security reviews found that Claude Code collects timezone and proxy information to identify users accessing the service from China via virtual private network. Anthropic described the system as a March experiment intended to prevent unauthorized reseller access. Alibaba has characterized the accurate identification of its engineers' unauthorized access as a backdoor.

"This constitutes an unacceptable security risk to our engineering environment," a company spokesperson said. Asked to specify which element posed the threat — the timezone datum, the proxy address, or the confirmation that the engineers were in Hangzhou — the spokesperson did not respond.

Affected developers will transition to Qoder, Alibaba's internal AI coding platform. Multiple engineers, reached for comment, described Qoder as "fine" and confirmed they had continued using Claude Code via VPN through the date of publication.

Anthropic did not comment on whether a successor detection mechanism would monitor compliance with the ban.

"We are committed to protecting our employees from tools that collect data about our employees," the spokesperson said.