4

At the EU level, the general legal framework for the online intermediary liability regime was established under the ECD in 2000. It introduced harmonised rules which apply to all providers of information society services, commonly referred to as online intermediary services, defined as services that are normally provided for remuneration or as a part of the economic activity, at a distance, by electronic means for the processing and storage of data upon an individual request of their user. [12] Article 14 of the ECD provides a special safe harbour regime for hosting service providers which store and host information by and at the request of their users, such as online marketplaces, social media networks, VSPs, etc. Due to its very nature, hosting services are often prone to be contaminated with illegal content uploaded by their users and therefore are subject to stricter exemption rules than other types of online intermediaries such as conduit and cashing service providers. [13] Accordingly, these providers are exempted from liabilities for the illegal content on their services uploaded by a user as long as (i) it has no actual knowledge of its user’s illegal activity and is not aware of facts, and circumstances from which the illegal activities or information is apparent [14] and (ii) once it obtains such knowledge/awareness, it acts expeditiously to remove or disable access to the information. [15] This exemption is applicable only to those cases where the activity of the hosting service providers is deemed merely technical, automatic and passive which implies that the online intermediary has neither knowledge of nor control over the information which is transmitted or stored. [16]

5

This safe harbour regime is supplemented by Article 15(1) which prohibits member states from imposing general obligations on online intermediaries to monitor the information transmitted or stored on their services, or to actively look for facts or circumstances indicating illegal activity. According to the EC’s Communication ‘A European Initiative in Electronic Commerce’ in 1997 [17], the European Parliament Resolution on this Communication in 1998 [18] and the First Report on the application of the ECD in 2003, this safe harbour regime including the prohibition on general monitoring obligation was rested mainly on five reasons: (i) online intermediaries, while in their infancy, lacked the technical capacity to actively and accurately monitor the massive amount of information transmitted via their services, [19] (ii) such monitoring obligation was deemed unfair as it creates a burden on those acting as passive mere intermediaries, [20] (iii) a desire not to deter a developing online commerce industry in the EU with “over-regulation”, (iv) the risk of over-blocking of legitimate content, i.e. free flow of information within the single market, due to the false positives of automated system or due to the tendency to avoid liability [21], and (v) the risk of creating actual knowledge and awareness which would result from an illegal content that slipped away from general proactive monitoring. [22] On the other hand, both the EC’s first report as well as Recital 47 of the ECD note that this prohibition covers only the monitoring obligation in a general manner and does not include monitoring obligations in a specific case. Furthermore, it does not preclude national courts to order the online intermediary to prevent an infringement [23] nor member states to impose a duty of care to hosting service providers to detect and prevent certain types of illegal activities. [24]

6

Although, it is obvious that monitoring means the supervision of data traffic on the service, the ECD fails to provide guidance on the difference between the monitoring obligation “of a general manner” and “in a specific case”. [25] Since the general monitoring prohibition determines the permissible scope of the measures which can be imposed on online intermediaries against illegal content, this ambiguity would likely cause problems in practice. Given that hosting service providers can still be required to prevent a specific infringement or certain illegal activities under the ECD [26], it becomes important to determine the extent of such preventive measures. [27] In fact, for the prevention of illegal content, the most effective option [28] becomes the adoption of filtering systems that monitor content either before or very shortly after it has been posted by its user. [29] Due to this ambiguity, the question as how to distinguish prohibited general monitoring obligations from permissible monitoring obligations has been addressed by the CJEU under the several preliminary rulings.

7

L’Oréal v eBay is the first case in which the CJEU assessed the compatibility of a court injunction on an online marketplace to prevent the future infringement of the trademark rights by internet users. The CJEU found that such preventive injunction would require eBay to conduct “active monitoring of all the data of each of its customers in order to prevent any future infringement” of L’Oréal trademarks and thus constitute general monitoring.  [30] Instead, the CJEU suggested two measures: firstly, the suspension of the infringing users who sold the counterfeit L’Oréal products on the platform in order to prevent further infringements of L’Oréal’s rights by the same users and secondly, the adoption of user identification measures to identify real persons behind the user accounts infringing copyrights and thus to prevent those suspended infringers from operating on the same platform under different user accounts.. [31] The rationale of these suggestions is found in the analysis made by Advocate General (“AG”) Jääskinen who determined double requirements of identifications of infringed right and of the infringer as an appropriate limit of a preventive measure. He opined that an injunction requiring an online intermediary to only target an infringement of the same trademarks by the same users would be permissible under Article 15(1) ECD. [32] It is argued that the CJEU’s suggestions are based on this opinion since both measures require the collective application of the detection of infringement of the specific trademark and identification of specific users. [33] This means that monitoring in a “specific case” must be understood in the sense of a specific incident of infringement, i.e. infringement by the specific users, rather than in the sense of all incidents of the same trademark infringement. The latter is found to require active monitoring of all the data of each of eBay’s customers, which therefore violates the prohibition on general monitoring. [34]

8

This interpretation was later tested in both the Scarlet Extended v SABAM  [35] and the SABAM v Netlog cases [36] in which the CJEU discussed whether an injunction ordering a mere conduit provider and a hosting service provider, respectively, to implement a permanent filtering system to prevent infringement of specific copyright-protected works, i.e., those listed in the repertoire of the Belgian collecting society SABAM, complies with Article 15(1) ECD. The CJEU noted that the implementation of such filtering system needs three main functions: (i) to identify content that includes copyright-protected works within all the content moving through the service, (ii) to assess whether those works are used unlawfully; and, if so, (iii) blocking or removing access to the content containing such illegal use of copyright-protected works. [37] Considering these functions, the CJEU concluded that such filtering mechanism would eventually require the active observation of all information provided by all users and thus it would amount to general monitoring. [38]

9

This conclusion is in line with the L’Oréal judgement. Although the injunctions in both cases were targeted to specific content, i.e. L’Oréal’s trademarks and SABAM’s works, the CJEU considered the monitoring of all activity by all users as general monitoring regardless of whether such monitoring is targeting only the infringements of specific rights. This means, due to its basic working principle, i.e., monitoring all users’ content, all possible filtering measures would fall under this classic generality. In fact, this ratione materiae is also adopted by AG Villalón Cruz in the Scarlet Extended v SABAM. He stated that the implementation of filtering measures requires prior monitoring of all information and without prior monitoring, these filtering measures cannot succeed. [39] Similarly, in the McFadden v Sony Music case, an injunction requiring a mere conduit provider to examine all information transmitting through its internet connection services in order to prevent third parties from infringing the particular copyright-protected works of Sony is found incompatible with Article 15(1) as it would require monitoring of all information from all users. [40] According to Senftleben and Angelopoulos (2020) [41] and Kulk and Borgesius (2013) [42], the CJEU’s findings in all these cases suggest that the permissible specific monitoring under Article 15(1) would be a filter system targeting specific, pre-notified infringements within the content posted by a specific group from among all of an intermediary’s users who are pre-identified as likely to share infringing content.

10

In 2019, the CJEU introduced a significant addition to this interpretation and widened the scope of permissible specific monitoring in the Glawischnig-Piesczek v Facebook case. [43] The CJEU permitted an injunction ordering a hosting service provider, Facebook, to remove content containing identical or essentially unchanged defamatory content that was previously declared illegal by a national court, “irrespective of who requested the storage of that information” [44], on condition that clear instructions must be given to the provider on how to identify such content so that it would not have to adjudicate the legality of the content. [45] For instance, any content containing the plaintiff's picture alongside a combination of certain insulting words, which have the same meaning to those used in the defamatory content, were determined as equivalent content in this context by the Austrian court. This means that to prevent the recurrence of such defamatory content, the online intermediary does not have any option but to monitor all information uploaded by all users which was explicitly rejected by the CJEU in the previous McFadden, SABAM and L’Oréal cases.

11

The reason behind this widening approach seems to be the CJEU’s acknowledgement of the dynamic nature of the social network environment which allows swift flow of the same information among its users and thus making monitoring meaningless to focus on pre-identified users. [46] Therefore, this judgement changed the scope of general monitoring, at least for defamatory cases, by allowing the active observation of all information uploaded by each service user in order to prevent pre-identified infringements. [47] According to the CJEU, the defining character of prohibited general monitoring becomes the requirements for online intermediaries to carry out an independent legal assessment of the illegal nature of the content. [48]

12

This broad interpretation was supported and the permissible scope of monitoring was further extended to copyright infringements in the Petersons/Elsevier v Youtube/Cyando case. The CJEU was asked whether an injunction for the removal and prevention of copyright infringing content, which exposes its addressee to unduly court costs, can be imposed on online intermediaries even if they fulfil the conditions of the safe harbour rules for hosting service providers under Article 14(1). [49] The CJEU noted that such an injunction would amount to the general monitoring obligation as it may force online intermediaries, which want to avoid court expenses, to actively monitor all the content uploaded by their users to prevent any copyright infringement. [50] However, it is allowed to impose a pre-condition for such an injunction requiring rightsholders to notify the online intermediary of an infringement prior to the commencement of court proceedings, in order to allow online intermediary to take necessary measures to prevent those notified infringements from recurrence and thus avoid being the subject of an injunction and subsequently court costs would not constitute general monitoring obligation. [51]

13

Although, the CJEU did not settle its ruling with the previous interpretation in the Glawischnig-Piesczek case, with respect to targeting copyright infringements instead of defamatory content, the AG Saugmandsgaard Øe’s opinion provides a convincing reconciliation. Upon assessing the identical and equivalent content standard determined in the Glawischnig-Piesczek case in the context of copyright law, he concluded that identical content means the content that contains the exact use of the same copyright-protected work which was previously found to be infringing, whereas equivalent content includes identical files that use the same work in the same way but which may have been uploaded in a different format. [52] For instance, a video showing an entire movie in a smaller screen frame on YouTube without any additional contextual information would comply with this equivalent infringing use of copyright-protected work standard. [53]

14

Moreover, the CJEU did not explain how exactly an online intermediary that previously was informed of an infringement should prevent the recurrence of that infringement. However, by analogy with the permissible duty of care obligation to prevent certain types of illegal content under Recital 48 the ECD, the CJEU’s judgement can be interpreted as: online intermediaries may be forced to filter all information on their services to detect certain infringing content which is pre-identified by a national court in line with Saugmandsgaard Øe’s standards. In fact, before the CJEU approved the contested condition for the preventive injunction in YouTube/Cyando case, it reiterated from the SABAM judgements that “requiring a service provider to introduce, (‘…’) [a] system which entails general and permanent monitoring in order to prevent any future infringement of intellectual property rights were incompatible with Article 15(1)”. [54] The difference between the contested condition and this quotation is that monitoring in the former is limited with the pre-notified copyright infringement and its obligation starts upon the receipt of a notification while the injunction in SABAM cases requires monitoring of any infringements containing specific copyright-protected works which needs a contextual analysis from an online intermediary for an indefinite time. As explained in the foregoing paragraph, this conclusion also reconciles with the CJEU’s approval for monitoring obligation for specific defamatory content in Glawischnig-Piesczek case. [55]

15

In the very recent ruling of Poland v European Parliament and the Council of the European Union C-401/19 case (“Poland v Parliament and Council”) [56], the CJEU confirmed this conclusion by reiterating its interpretations of general monitoring in both YouTube/Cyando and Glawischnig-Piesczek cases. The CJEU was asked to annul Article 17(4) of Copyright Directive which provides for the obligation for online content sharing services, a type of hosting service providers, to make their best effort, with high industry standards of professional diligence, to prevent the occurrence of a copyright infringement if the service providers concerned have received from the rightsholders sufficiently substantiated, relevant and necessary information of specific copyright infringement. First of all, the court concluded that requirement of best effort with high industry standards of professional diligence to prevent the occurrence of a copyright infringement obliges very large content sharing services, which receive thousands or millions of daily uploads, to carry out prior review and filtering of online content via automatic recognition and filtering tools. [57] However, the court also notes that this obligation becomes applicable only after the service provider receives sufficiently substantiated notice the specific infringement or relevant and necessary information regarding the copy-right protected work which must enable the service provider to identify the unlawful content without conducting legal assessment. [58] Lastly, once again the court pointed out that generally the service providers “cannot be required to prevent the uploading and making available to the public of content which, in order to be found unlawful, would require an independent assessment of the content by them in the light of the information provided by the rightholders and of any exceptions and limitations to copyright (‘…’)” as this leads to general monitoring obligation. [59]

16

The CJEU’s recent clarification of general monitoring obligation confirms that any obligation to online intermediaries requiring filtering all the information on their services to detect and remove the illegal content on condition that the identification of such content must not require “an independent assessment” or “legal examination”. This means online intermediaries should not be required, for example, to carry out a contextual analysis of content that contains the defamatory content pre-identified by a court but in a significantly different context or which includes a copyright protected work used in such a way that contrast the information provided by rightsholders with applicable copyright exceptions. [60] In line with Saugmandsgaard Øe’s opinion in both the Poland v Parliament and Council case and the Youtube/Cyando case, the CJEU seems to agree that any obligation to implement upload filters against manifestly illegal content, the illegal nature of which either is clear and obvious to a reasonable person or has been previously determined by a court, does not constitute general monitoring obligation. [61]

17

Within its interpretative case-law, the CJEU noted that while the monitoring obligations generally aim to protect the rights and interests of the people, e.g. the right to intellectual property [62], the right to reputation [63] from the infringements by internet users, it also burdens the internet users’ rights to privacy and data protection, freedom of expression and information and the online intermediary’s freedom to conduct a business under Articles 8, 11, and 16 of the Charter respectively. [64] In the face of this clash, the CJEU developed a fair balance test to strike the balance between these competing rights and interests within the framework of the online intermediary liability regime. The analysis of the CJEU’s fair balance test would present how permissible specific monitoring obligations can be implemented. The justification for the imposition of liability on online intermediaries is supplemented by the context of the rulings of the European Court of Human Rights in which freedom of expression and information are balanced against the right to privacy in reputation. [65]

18

Apart from these fundamental rights, scholars have also raised concerns over the negative impacts of automated monitoring systems on the internet users’ rights to equality and non-discrimination due to inherent bias in algorithms and thus the absence of the rights to a fair trial and effective remedy of those whose online expression is restricted by the over-blocking. [66] As both of these issues are discussed by the CJEU and AGs in relation to the risk of over-blocking of the users’ legitimate expressions, this section will evaluate the impact on these two fundamental rights under the CJEU’s fair balance test for freedom of expression and information and then the suitability of the balancing approach in the context of general monitoring of online content will be questioned below.

41

This article shows that, at the beginning of the last decade, the CJEU adopted a broad interpretation of the concept of general monitoring. Accordingly, the imposition of any obligation requiring an online intermediary to monitor all information of all service users to filter infringements falls within the scope of the prohibition as it constitutes an excessive burden on the fundamental rights of online intermediaries as well as of internet users. [125] However, in the recent cases, the CJEU has recognised the difficulties for the targeting specific infringement from particular users due to fast-paced information flow of the internet realm, and acknowledged the fact that any preventive measure against illegal content cannot be effective without prior monitoring of all information flowing through the service. [126]

42

Perhaps, this shift from banning monitoring of all information to allowing the same practices for specific infringements can be explained by assessing the validity of the reasons behind the adaptation of the prohibition on the general monitoring obligation in the ECD at the beginning of this millennium. [127] Given the advancement in technology and the rapid economic growth of online intermediaries in recent years, the reasons for the lack of technical capacity and the desire not to deter a developing industry seem to have lost their validity in the eyes of the CJEU. Furthermore, the risk of creating actual knowledge and awareness by monitoring all the content including illegal but not notified ones has also been refused by the CJEU in the YouTube/Cyando case. [128] On the other hand, the risk of over-blocking and the unfairness of imposing obligation upon those mere intermediaries seem to be only valid reasons behind the CJEU’s interpretation of Article 15(1) of the ECD. In relation to these concerns, both the European Courts seem to limit the scope of proactive measures against manifestly illegal content that would not require the online intermediary to conduct any legal assessment and only allow its imposition on financially and technically resourceful intermediaries that have influence over the curation of content instead of merely hosting them. [129] Lastly, in any circumstance, intermediaries must implement effective redress mechanisms and safeguards for legitimate personal data processing for internet users. [130]

43

Overall, as per the CJEU’s case-law, the permissible monitoring obligations must: (i) be targeted to online content [131] which has been previously identified as illegal by a court [132] or which is manifestly illegal for a reasonable person [133], (ii) not require an additional independent legal assessment to identify, [134] (iii) be effective [135], reasonable [136] and propriate in accordance with the technical, operational and financial capabilities of the intermediary, [137] and with the impact of illegal content [138], for instance anyone of the GAFAM platforms [139], (iv) be carried out on the legitimate basis for the processing of personal data [140], and (v) be supplemented with an appropriate redress mechanism granted to internet users [141].

44

It must be noted that Senftleben and Angelopoulos (2021) refused this general conclusion as they believe that the standards for the prohibition on general monitoring must be “specific in respect of both the protected subject matter and potential infringers”. [142] First, they argued that the Glawischnig-Piesczek judgement is incompatible with the CJEU’s rulings in the SABAM, McFadden, and L’Oréal cases because the infringements in intellectual property law depend not only on the specific use of work but also on the identity of the specific group of users. [143] Furthermore, Senftleben and Angelopoulos (2021) also pointed out that while it is often sufficient to identify the protected work that is fixed after the first publication in copyright issues, defamation cases, by contrast, depend on the nature of uploaded content and the use of specific defamatory elements in specific contexts. [144] Due to these substantial differences, the standards of general monitoring will be shaped based on “the nature and scope of the legal position, in respect of which the imposition of duties of care, including the introduction of content moderation duties, is requested”. [145] While this is a plausible argument, considering the horizontal nature of the ECD, and both AG Saugmandsgaard-ØE’s interpretation of manifestly illegal content in Poland v Parliament and Council case with the explicit reference to AG Spunzar’s interpretation in Glawischnig-Piesczek case, [146] this article accepts the horizontal effect of the CJEU’s interpretation of the scope of specific monitoring in line with Reda et al (2020), and Van Eecke (2011), and it argues that monitoring obligation for specific infringement is permissible regardless of the nature of infringement as long as the identification of illegal content can be carried out without any legal assessment of intermediaries and the effective redress mechanisms are implemented. [147]

46

In 2018, the EU amended the AVMSD [148] to introduce new requirements for VSP provider, a recently defined subset of hosting service provider. [149] According to Article 28b, member states must ensure that VSP providers adopt “appropriate, practicable and proportionate” measures to protect minors from online content which may impair their physical, mental or moral development and the general public from the dissemination of content containing hate speech and incitement to violence, provocation to commit terrorist offence, child sexual abuse material and racism and xenophobia. [150] The AVMSD further provides a non-exhaustive list of measures that are deemed appropriate by EU legislators including, among others, the notice and take down systems based on user’s reporting [151] but also allow Member States to impose more detailed and stricter measures on VSPs. [152] However, these measures shall not lead to ex-ante control measures or upload-filtering of content, which do not comply with Article 15 ECD. [153]

47

It becomes evident that the EU lawmakers consider the balance test requirement for the determination of appropriate, practicable and proportionate measures in line with the CJEU’s case-law. [154] But, it is not clear as to what measures could be stricter than a notice and takedown procedure in the context of the available technology [155] but do not constitute ex-ante content moderation measures, which are clearly considered a violation of Article 15 of the ECD. Moreover, this prohibition of ex-ante control measures and upload-filters fails to reconcile the YouTube/Cynado, Glawishking and Poland v Parliament/Council rulings as well as the ECtHR’s case-law, which allow the imposition of monitoring obligation to prevent manifestly illegal content. [156]

48

The Regulation on Preventing Dissemination of Terrorist Content Online enacted in May 2021 imposes obligations on hosting service providers to remove terrorist content within an hour upon receipt of a notification from a competent authority. [157]Hosting service providers must take specific measures to protect their services from being misused for the dissemination of terrorist content if the competent authority finds the service is exposed to terrorist content on basis of certain factors, such as having received two or more removal orders from a competent authority within the past twelve (12) months. In line with the settled balancing test of the CJEU, the Regulation also grants freedom to hosting service providers on their choice of specific measures on condition that these measures must be effective in mitigating the risk, proportionate with the technical, financial, and operational capabilities, the number of users of the hosting service provider and the amount of content they provide. [158] The competent authorities also have power to require additional specific measures if they find the hosting service provider’s measures are insufficient to address the risks. [159] Nevertheless, the imposition of any requirement leading a general obligation to monitor or actively seek facts or circumstances indicating illegal activity under Article 15(1) ECD or to use of automated tools by hosting providers are prohibited. [160]

49

Once again, the prohibition on general monitoring appears as the borderline for statutory specific measures. However, the whole system established under the Regulation including the obligation to remove notified terrorist content and to take specific measures for the protection of the service, seems to give no other option to hosting providers but to take certain proactive measures in practice. First, as the obligation to take specific measures is triggered by the receipt of more than two removal orders,without any preventive measures, the previously removed terrorist content can be easily re-uploaded and cause the competent authorities to issue additional third removal order which eventually triggers the requirement to take so-called specific measures. Moreover, Article 5(2)(a) classifies “appropriate technical and operational measures or capacities, such as appropriate staffing or technical means to identify and expeditiously remove or disable access to terrorist content” as a permissible specific measure which clearly requires de facto monitoring of uploaded content in order to identify terrorist content. Given that Article 7(3)(b) also requires providers to publish information about measures taken to address the reappearance of previously removed content annually, it becomes apparent that hosting providers are expected to take preventive measure at some degree, for instance against the pre-identified terrorist content. [161]

50

To a certain extent, the suspension of users or accounts that are identified as terrorist content uploader can be considered a preventive measure. However, the privacy concerns over the loss of online anonymity [162] and the availability of technologies that provide anonymity [163] would hamper the effectiveness of these suspensions. Therefore, considering the requirement for specific measures to be effective, appropriate and proportionate in accordance with a hosting provider’s size, technical and economic capacity, and the number of its users, [164] it becomes evident that major hosting service providers enabling access to user content in large scales do not have any other option to effectively mitigate the dissemination of terrorist content but to implement the filtering measure for pre-identified terrorist content. Indeed, this understanding would comply not only with both the EU legislators’ recent statements concerning the proactive measures against manifestly illegal content [165] but also with the CJEU’s interpretation of Article 15(1) ECD. [166] In fact, all necessary safeguards for fundamental rights stipulated by the CJEU have already been considered under the Terrorist Content Regulation such as the proportionality test, human oversight, and verification in the use of automated tools against over-blocking and the introduction of complaint and redress mechanisms. [167] However, if the CJEU’s interpretation is accepted and the hosting providers can be forced to take technical and operational measures to identify and expeditiously remove pre-identified terrorist content under Article 5(2)(a), this Regulation would be incompatible with the prohibition of ex-ante control measures as provided for under the AVMSD.

51

The obligations stipulated under Article 17(4) of the Copyright Directive were the subject of one of the most influential CJEU rulings, Poland v Parliament/Council. According to Article 17(4), online content-sharing service providers (“OCSSP”) [168] which have not obtained an authorization from the rightholders must demonstrate that they have: (i) made best efforts to ensure the unavailability of specific copyright-protected works for which the relevant rightholders must have provided the OCSSP with the relevant and necessary information and (ii) acted expeditiously, upon receiving a sufficiently substantiated notice from the rightholders, to remove the content infringing the notified works and made best offers to prevent their future uploads. [169] The assessment of “best efforts” is made in accordance with “high industry standards of professional diligence” and the principle of proportionality with regard to the number of elements such as the type, the audience and the size of the service, the evolving state of the art to avoid the availability of different types of content and the cost of such means for the services. [170] In fact, Article 17(6) exempts new OCSSPs from the “best effort” requirement that have been active in the EU for less than 3 years, have less than 5 million monthly unique visitors and have an annual turnover of less than 10 million euros. Lastly, Article 17(8) explicitly states that the application of best effort requirements under Article 17 shall not lead to any general monitoring obligation. [171]

52

The EC’s Guidance on Article 17 recognises the content recognition technologies as a method “commonly used today to manage the use of copyright protected content, at least by the major online content-sharing service providers and as regards certain types of content” and note that these technologies can be considered as the market standards to filter and block manifestly infringing content for large OSSPs. [172] In the Poland v Parliament/Council case, the CJEU confirmed this position by explicitly announcing that the requirement for use of automated recognition and filtering technologies under the best effort obligations, do not amount to general monitoring obligations that could hamper the providers' freedom to conduct a business. [173] Basically, both the EC and CJEU agreed that upload filters can be compatible with the prohibition as long as the scope of filtering measures is limited to specific infringement identified by courts or rightholders and which is specific enough to be detected by automated tools. [174] In addition, they noted that certain safeguards must be implemented for fundamental rights in particularly freedom of expression and right to remedy. [175]

61

It is evident that there is a lack of a uniform application of general monitoring prohibition within the EU intermediary liability regime. Whereas the AVMSD qualifies ex-ante control measures and upload-filters as prohibited general monitoring regardless of the nature of the content and the Terrorist Content Regulation prohibits obligation to use automated tools against terrorist content [198], the Copyright Directive, in line with the CJEU’s interpretation, allows obligation to implement automated filtering measures by VSPs against specific copyright infringements. In practice, these different approaches towards content moderation measures might cause VSPs which host both video, image and textual content like Instagram or Twitter to face potential difficulties depending on whether manifestly illegal content is posted by video, or within a still image or as a text. [199]

62

Firstly, while both the AVMSD and the Terrorist Content Regulation aim to tackle with the dissemination of the terrorist content, the required measures differ significantly under each legislation. According to AVMSD, VSPs cannot be forced to implement upload-filters, but on the other hand, the Terrorist Content Regulation expects them to prevent the recurrence of previously removed terrorist content. This means VSPs cannot be required to take measures to “identify and expeditiously remove” pre-identified terrorist content under Article 5(2)(a) the Terrorist Content Regulation, as it contradicts with the AVMSD. However, without implementing ex-ante monitoring to tackle previously removed terrorist content and by solely relying on reactive measures, there will be a loop of constant uploads and one-hour removals of the same terrorist content.

63

If the prohibition of general monitoring under the Terrorist Content Regulation is read in compliance with the AVMSD, then the specific measure under Article 5 would not go beyond being the supplementary list to the non-exhaustive list of appropriate measures under Article 28b of the AVMSD. Furthermore, there is nothing to stop national courts from issuing an injunction on VSPs to prevent pre-identified terrorist content on its service. This will eventually lead to VPSs with thousands of daily uploads to implement an automated filtering system to comply with such injunction even if it cannot be required under the Terrorist Content Regulation. Given that the CJEU identified automated recognition and filtering tools as an effective measure against dissemination of illegal content in the Poland v Parliament/Council judgement, the prohibition on requirement for the use of automated tools under the Terrorist Content Regulation perhaps becomes an empty shell in practice at least for large service providers as they do not have any other option but to implement automated systems other than employing thousands of human moderators.

64

Secondly, there is an imbalance between rights and interests under the current legislative framework. The interests at stake for the prevention of reappearance of content containing non-consensual sexual videos, child sexual abuse, provocation to commit a terrorist or extremist offence which are pre-identified by judicial authorities as illegal are considerably higher than the interest of copyright holders protected under Article 17 of the Copyright Directive. [200] In fact, today, the automatic duplicate-detection systems are the most commonly deployed systems to filter out duplicates of known, specific terrorist or child exploitation images, audio, or videos in practice, and they even provide more successful results than human moderators. [201] On the other hand, despite the recent developments in the content recognition technologies, empirical studies show that these systems still perform poorly for the detection of infringements that contain the same, previously notified copyright-protected work. [202] Considering the balance test conducted by both European Courts, it is evident that the monitoring obligations against these manifestly illegal content would constitute a more proportionate limitation on the exercise of the freedom of expression and VPS’s freedom to conduct a business. Therefore, the AVMSD’s interpretation of general monitoring which covers upload-filters against child sexual abuse and provocation to terrorism and extremism seriously hamper the EU’s aim to create a safe digital single market. [203]

65

On the other hand, this article does not disregard the concerns over the risk of excessive restriction of fundamental rights posed by any monitoring obligation requiring an independent assessment of VPSs or national administrative authorities. In addition to risks explained under the case-law review above, as per Balkin (2014), by imposing general monitoring obligation, governments can acquire the power to impose “collateral censorship” on free speech online through the hands of the VSPs. [204] Moreover, it is evident that monitoring of all user information to detect not only manifestly illegal but also other types of illegal content, which require legal assessment, would preclude individuals from sharing and discussing their ideas online and eventually harm their intellectual privacy. [205] Unfortunately, it seems like these considerations are overlooked under the balancing exercise of the European human rights adjudicators. Whereas the monitoring obligation should be accepted as a direct interference with the freedom of expression, the right to privacy of internet users and the proportionality test as per Article 52 of the Charter needs to be applied in this context, we have witnessed mere utilitarian comparison of fundamental rights without considering principled hierarchy of interests. As rightly pointed out by van der Sloot (1998), the case law review under Chapter 2 Section 3 of this paper presents minor interests, “such as not being called a rascal or the copyright protection of a commercial business,” are promoted to fundamental rights discourse under the balancing exercise. [206] The special protection to certain principles and interests which are deemed essential not only to human life but also to modern day democratic societies, such as the right to receive and impart information on VSPs seems to be forgotten in the interpretation of general monitoring obligations by both the CJEU and the EU legislators.

66

Lastly, imposition of specific monitoring obligations against defamatory content might be very problematic in practice. While the copyrighted works can be detected by automatic recognition tools regardless of the format of content, such as video, text, audio, automatic recognition of defamatory content may not always be easily done. For instance, if the defamatory content in Glawischnig-Piesczek case would have been re-uploaded as a video to Facebook where a random person reads the text of the original defamatory content in a different but commonly used language, would this video still qualify as equivalent content to original as the message remains essentially unaltered? Or what if this video does not include any audio but just shows series of cardboards where the original messages are written? Or should we expect VSPs to prevent occurrence of such video if it was a part of reporting activities of an amateur journalist?

67

In light of these considerations, it becomes evident that one horizontally applicable prohibition not only creates legal uncertainty for VSPs but also fails to address interests of internet users in the online realm. Therefore, at the current situation, the evolving content/sector-specific EU legislations may include provisions which clearly distinguish the default prohibition on general monitoring obligations from the context-specific measures and which are tailored to address each specific type of illegality in a limited scope and under conditions that overwhelmingly safeguard the interest of individuals in exercising their freedom of expression and right to privacy. [207]