Alibaba has banned the use of Anthropic’s Claude Code across its engineering divisions following reports that the tool contained covert anti-distillation logic targeting specific geographic IP ranges.

Key Points

• Reports indicate Claude Code used hidden system prompts and IP filtering to detect and disrupt potential model extraction attempts.
• Alibaba’s security team flagged the tool for unauthorized data transmission and opaque behavior that prevented internal auditing.
• Anthropic reportedly acknowledged the feature as an intellectual property protection measure and pledged to remove it in a future update.
• The incident originated from community-led reverse engineering, which identified anomalous network traffic and hidden instructions within the software.
• Other technology firms are reportedly reviewing their AI tool deployments, though no further public bans have been confirmed.

Why it Matters

This incident highlights a growing tension between AI developers protecting proprietary models and the enterprise need for transparent, auditable software. It underscores the necessity for organizations to rigorously vet AI coding assistants, as covert features can create significant security and compliance risks for sensitive corporate environments.