Deepfake detection ranks first among the technologies that JRC and Europol experts consider urgent for European police forces

Artificially generated images, voices, and video recordings are becoming convincing enough to be used in fraud, identity theft, blackmail, disinformation, and the fabrication of materials that may end up in an investigation. In a joint technology foresight exercise, experts consulted by the Joint Research Centre and Europol ranked deepfake detection at the top of the technologies requiring rapid solutions and new digital forensics capabilities.

Deepfakes have moved beyond the stage where they were primarily a problem of false content distributed online and are becoming a direct challenge for criminal investigations. In a joint report produced by the European Commission’s Joint Research Centre and Europol, the experts consulted ranked deepfake detection first among the emerging technologies that European law enforcement authorities should monitor and for which solutions are needed in the near term. The risks identified range from impersonation and identity theft to financial fraud, blackmail, manipulation of public opinion, and the creation of images or recordings that can be presented as authentic evidence in an investigation.

1. Experts consulted by the JRC and Europol considered deepfake detection the topic with the greatest impact and highest urgency among the technologies selected in the foresight exercise.

2. The report warns that deepfakes can facilitate identity theft, blackmail, fraud, disinformation, and the falsification of digital materials relevant to investigations.

3. Detection increasingly relies on artificial intelligence, analysis of digital traces left by manipulation, and verification methods that go beyond what an investigator can observe directly.

4. Neural networks, Vision Transformers, forensic noise trace analysis, and watermarking methods are being examined to identify changes made to images and videos.

5. The report recommends investment in digital forensics expertise, staff training, and cooperation with researchers and technology companies, without turning the exercise’s conclusions into a new EU operational obligation.

The report “Horizon scanning on emerging privacy enhancement technologies” was produced by the EU Policy Lab within the Joint Research Centre and Europol’s Innovation Lab to anticipate technological changes that may influence police work in the coming years. The team began with more than 200 signals concerning technologies and factors that could accelerate or slow their development, then selected 178 for a workshop attended by approximately 20 specialists from policing, research, computer science, law, public policy, business, and civil society. This selection resulted in ten technologies and areas considered particularly relevant to the future work of authorities, with deepfake detection receiving the strongest assessment in terms of impact and urgency.

The stakes arise from the speed at which synthetic content is changing. Deepfakes can reproduce a person’s face, voice, and behavior and create materials in which that person appears to have said or done things that never happened. As generation tools become more accessible and their results harder to recognize visually, verifying authenticity can no longer depend on the experience or intuition of someone looking at an image or listening to a recording.

For police, the consequences are far more concrete than the mere circulation of forgeries on the internet. Synthetic material can be used to assume someone’s identity, authorize or simulate a transaction, construct a fraud attempt, apply pressure through blackmail, or produce a fake that imitates digital evidence. The report explicitly mentions the possibility that deepfakes could undermine legal proceedings and aggravate existing forms of cybercrime.

This creates an additional problem for investigators and courts: establishing the authenticity of material is becoming an increasingly important part of evaluating evidence. An image or video recording may still have significant evidentiary value, but the origin, file history, and technical traces of possible manipulation carry greater weight in an environment where synthetic content is becoming more realistic.

Deepfakes can also affect trust in authentic materials. A person confronted with a genuine recording may try to present it as AI-generated, while investigators need tools robust enough to demonstrate what is original and what has been altered. The report treats this development as a challenge to the quality of evidence and to public trust in legal and institutional processes.

Detection technologies are evolving in parallel with generation tools. One category already in use is convolutional neural networks, deep-learning models that look for anomalies in textures, edges, expressions, or other characteristics of images. The report also mentions combinations of Vision Transformers and EfficientNet B0, which seek to achieve better results by analyzing an image’s global and local features simultaneously.

Another approach focuses on the technical traces that the generation process may leave in files. Forensic noise trace analysis looks for subtle digital noise patterns that are not visible to the human eye but may indicate the intervention of a generation model or manipulation technique. Real-time detection solutions are also being tested for situations in which verification must be performed very quickly.

The report also examines prevention through watermarking. Under this approach, certain characteristics of facial identity can receive digital markers that enable subsequent verification of integrity and can signal that an image has been altered. Such solutions seek to shift part of the effort from discovering a forgery after distribution toward authenticating material from the moment it is created or recorded.

None of the methods analyzed is presented as a definitive solution. Generation models are evolving rapidly, and a detector effective against one type of deepfake may perform less well against a new technique. One important area of research is precisely the ability of detectors to recognize manipulations they did not encounter in the data used for training.

For authorities, this development involves more than acquiring software. The report emphasizes the need for forensic expertise, personnel trained to understand the limitations of algorithms, and procedures through which technical conclusions can be explained and documented in a criminal case file. A probability provided by an AI system is much more useful in an investigation when it can be supported by an intelligible and verifiable technical analysis.

This requirement is connected to the use of AI in evidence and expert assessments. Models that operate as “black boxes” can create problems when a prosecutor, lawyer, or judge needs to understand how the conclusion that material was falsified was reached. For digital forensics, technical performance and explainability thus become two conditions that must be developed together.

The JRC and Europol regard cooperation among authorities, researchers, and technology companies as one of the conditions for keeping pace. Generation and detection systems change faster than the usual procurement and training cycles in the public sector, and an institution’s ability to verify deepfakes depends on access to expertise and the continuous updating of tools.

The report also approaches deepfakes from the perspective of identity. Control over a person’s image and voice is linked to privacy and personal security, while impersonation can simultaneously affect the person targeted, the institutions trying to verify their identity, and third parties who receive the false material.

The issue is particularly relevant in financial and legal transactions, where deepfakes can be used to simulate a person’s identity or create a false context around an operation. The authors also include the risk of identity theft, one of the forms of crime that may become easier to automate and scale as AI tools improve.

The report does not provide a statistic on the current number of European investigations affected by deepfakes, nor does it say that they have become the main form of digital crime in the EU. Their first-place ranking reflects participants’ assessment of technological impact and urgency and indicates the area in which authorities should prepare their capabilities in advance.

This distinction is also important for how the publication should be read. The study is a foresight exercise, based primarily on a workshop and a limited number of interviews, and the authors acknowledge that some conclusions require further research. Its role is to show where problems and opportunities may arise that are important enough to justify preparation before they reach a large scale.

The Joint Research Centre is the European Commission’s scientific service, while Europol’s Innovation Lab monitors technologies that may change the work of law enforcement authorities. The joint report published in 2026 examines these technologies from two perspectives: their ability to protect data and assist investigations, and the possibility that the same tools could be used to conceal or facilitate criminal activities.

Deepfake detection opens the series of technologies analyzed in the report and received the strongest urgency rating in the workshop. Other chapters separately address post-quantum cryptography, computing on encrypted data, AI that protects data, 6G, and other emerging technologies, allowing each topic to be analyzed without overlapping the conclusions between them.

https://2eu.brussels/ro/news/detectarea-deepfake-urilor-pe-primul-loc-intre-tehnologiile-pe-care-expertii-jrc-si-europol-le-considera-urgente-pentru-politiile-europene