• Meta and Pinterest denied French regulator Arcom access to test their content recognition tools.
• Arcom partnered with rights holders to evaluate Meta’s Rights Manager, revealing inconsistencies.

Meta has refused to grant French telecommunications regulator Arcom direct access to its Rights Manager tool, exposing a gap in regulatory oversight of platforms’ content moderation systems. The social media giant’s decision forced the authority to pursue alternative testing methods, which highlighted concerns about the accountability of automated content recognition systems, particularly with respect to static images on social media platforms.

The controversy stems from the implementation of Article 17 of the EU Copyright Directive, which requires content-sharing platforms to ensure uploaded content doesn’t infringe on copyright. Following a 2023 consultation that revealed photographers’ dissatisfaction with existing tools, Arcom sought to evaluate the effectiveness of content recognition systems, particularly those handling static images.

Meta’s rejection of Arcom’s request was particularly notable given the regulator’s legal authority under Article L331-18 of the CPI to assess such measures. The tech giant and Pinterest justified their refusal by stating that their tools are exclusively reserved for rights holders – a status Arcom doesn’t hold. 

Pinterest’s situation proved more problematic than Meta’s, as Arcom couldn’t find a single rightsholder using its Claims Portal tool to facilitate testing of their copyright material. Arcom adapted its approach by partnering with existing rights holders to conduct tests on Meta’s Rights Manager. 

The results revealed significant inconsistencies in the tool’s performance across different scenarios. While the system successfully identified images posted in public spaces and from private Instagram profiles, it failed to detect content in several key areas, including:

• Public Facebook albums,
• Facebook group posts (both public and private),
• Posts with restricted audience settings,
• Some previously restricted posts, even after they were made public.

The investigation also exposed concerns about limitations in the Rights Manager’s ability to detect modified images. While the tool showed resilience against certain modifications, it struggled with common alterations that minimally impact visual recognition, such as mirror effects, saturation adjustments, and cropping. The cropping issue proved particularly problematic on Instagram, where the platform’s interface often requires users to crop images to meet format requirements.

Meta has acknowledged the inconsistencies, stating that it was previously unaware of some issues and is conducting internal investigations. However, the company’s initial reluctance to allow direct testing raises questions about transparency and accountability in content moderation systems.

The situation highlights a growing tension between regulatory oversight and platform autonomy. While social media companies have developed sophisticated content recognition tools to comply with copyright laws, their effectiveness still needs to be improved to satisfy independent verifiers. A lack of transparency and effectiveness will undermine creators’ rights and don’t bode well for the emergence of fair content moderation practices.

The standoff between Meta and Arcom could have broader implications for digital regulation in the EU. As content-sharing platforms continue to rely heavily on automated systems for copyright-infringement enforcement, regulators’ ability to independently verify the effectiveness of those automated measures becomes increasingly crucial. 

This incident may prompt discussions about whether platforms should be required to provide regulators with direct access to their content moderation tools, regardless of rightsholder status. For creators and users alike, the findings underscore the current limitations of automated content recognition systems and the need for more robust and transparent solutions. 

As the volume of user-generated content continues to grow, the effectiveness of these tools – and platforms’ willingness to submit them to independent testing – will remain cental in the debate about copyright protection.