Challenges and limitations of ACR technologies to protect IP rights

Challenges and limitations of ACR technologies to protect IP rights

Amin Hosseini is an IPilogue Writer and an LLM Candidate at Osgood Hall Law School.

In recent years, many technical mechanisms to protect intellectual property have given their helping hands to IP owners and technology developers to enhance IP management and protection. ACR technology is a prominent example.

Automatic content recognition (ACR) refers to the ability of an application to identify content based on sampling a portion and comparing it with a source service. ACR technologies are potential tools that can increase protection of IP rights through several specific use cases. The European Union Intellectual Office (EUIPO) published a discussion paper, the second stage of a project begun in 2019, titled "Automated Content Recognition: IP Enforcement and management use cases" on September 22, 2022, and thoroughly examined the use cases, which are as follows:

  1. Solutions to detect IP-infringing listings on e-commerce marketplaces;
  2. Smartphone solutions to detect genuine or counterfeit products;
  3. Solutions to recognize 3D printing files and 3D-printed products;
  4. Solutions to protect and manage copyright and neighbouring rights on content-sharing services; and
  5. Solutions to identify live streams of IP-protected content.

The analysis for each use case includes a description of the problems that ACR solutions can solve, as well as the benefits and drawbacks, not only from a technical standpoint but also in dealing with complex legal issues.

Data is indispensable in training recognition solutions to identify infringement on online marketplaces. To ensure accuracy, content recognition technologies require high-quality data and computational resources. E-commerce marketplaces need many photos, descriptions, and other metadata to create, train, and use recognition technologies. In addition to being a costly process, it is challenging to determine what legally constitutes an infringement of IP. That is why human experts must evaluate the infringement in question.

For smartphones, where they perform to detect genuine or counterfeit products, the quality of the datasets is critical in the training and improvement of the detection models since it allows the ACR technologies to analyze and compare the data gathered and received. This is done by fingerprinting, which is conducted by recognizing an extract of a piece of content. Data collection, storage, and processing necessitate  numerous computer resources when feeding databases with captured photographs produced by smartphone users. Resources required to create, store, and compare fingerprints might vary significantly depending on the product and the fingerprint's complexity.

The scenario is different for 3D files and products. The use case employs watermarking software to detect IP infringement. Since there is no requirement for a reference database, it is not necessary to use a lot of data or processing power to recognize previously watermarked files or printed products. However, it can take a lot of resources to add watermarks to each  copy of a file or printed product and then detect them once more. In addition, without standardized watermarking methods, a system employing one technology cannot read a watermark produced by a system employing a different technology.

Most content-sharing services rely on fingerprinting-based solutions offered by third-party vendors. Hence, it is exceedingly challenging to locate reference datasets for developing ACR systems because not all businesses use the same test environments. Furthermore, based on article 17.4 of directive (EU) 2019/790, for content-sharing service providers to take appropriate action and make sure that specific copyright-protected content is not accessible on their platform, IP owners must provide them with all relevant and necessary information. The effectiveness of this mechanism will depend on how this provision is put into practice and how well the relevant parties work together.

Cases where ACR systems incorrectly match the content supplied to a content-sharing service with content in the reference database and inappropriately restricted it is also a significant legal challenge, especially when they impair users' fundamental freedoms.

According to the EUIPO`s paper, another issue is the lack of standards and interoperability between various fingerprinting-based solutions, which forces IP owners to submit their materials or related fingerprints to various content-sharing services and solution providers without the benefit of specific procedures.

A further challenge for ACR technologies in content-sharing platforms is applying specific copyright limitations or exceptions to content.  A human review procedure where IP owners and users can voice their opinions may be necessary. The removal of perfectly lawful content can also be a problem if someone makes erroneous or abusive claims of ownership over it.

Finally, feeding the content into the ACR reference databases to discover unauthorized streams is a problem in the context of securing live-streamed events, especially with fingerprinting solutions. There needs to be further agreements and technological arrangements between IP owners and ACR systems.

Further Reading

For more information on Artificial Intelligence & Intellectual Property see Daniel Kiat Boon, Detecting and Prosecuting IP Infringement with AI: Can the AI Genie Repulse the Forty Counterfeit Thieves of Alibaba? (November 29, 2019). Artificial Intelligence & Intellectual Property, 2019, Available at SSRN: