Anthropic accuses rivals of Claude distillation

This is less about a single abuse report than a bid to redefine distillation as a strategic security issue, not just a terms-of-service violation.

• –Anthropic is arguing that stolen model behavior includes stripped-down access to high-value capabilities like reasoning traces, agent workflows, and coding performance
• –The company ties distillation directly to export controls, making the case that model copying can blunt U.S. chip-policy advantages without matching frontier R&D from scratch
• –For AI developers, the important subtext is that labs are likely to harden APIs, tighten account verification, and get more aggressive about detecting synthetic traffic patterns
• –The framing matters because it raises the question of whether this is pure security disclosure, competitive narrative-setting, or both
• –The broader industry signal is clear: frontier labs increasingly see model outputs themselves as defendable infrastructure, not just monetizable API traffic