I believe it would come down to proving malicious intent on behalf of the user, and from there it would become a question of "does Clawdbot reliably produce malicious code, and if so, is it a threat to anyone besides the user?"