Deepfake detection technology

This publication is licensed under the terms of the Open Government Licence v3.0 except where otherwise stated. To view this licence, visit nationalarchives.gov.uk/doc/open-government-licence/version/3 or write to the Information Policy Team, The National Archives, Kew, London TW9 4DU, or email: [email protected] .

Where we have identified any third party copyright information you will need to obtain permission from the copyright holders concerned.

This publication is available at

Deepfake detection technology has emerged as a critical field in recent years, combating the growing threats posed by deepfakes, a subset of synthetic media involving audio-visual content that is generated or manipulated using AI to misrepresent someone or something.

Despite progress being made through initiatives such as the UK government’s Deepfake Detection Challenge and ongoing research efforts in AI identification, particularly within the policing and law enforcement sectors, the development of deepfake detection technology remains in its early stages, and the market itself remains nascent.

In late 2024, the Department for Science, Innovation and Technology (DSIT) commissioned PUBLIC to investigate the state of the UK deepfake detection market and its potential for growth. Through a combination of a literature review, expert engagement, and provider mapping, this report aims to establish a robust evidence base to inform how the government can support the continued growth of this market.

This report was prepared by a specialist digital advisory firm, PUBLIC Group International Ltd, who have been supporting DSIT with this analysis. This report is not official government policy.

The mitigation of deepfake threats occurs at multiple stages: before, during and after their creation. This report specifically focuses on deepfake detection, which encompasses a range of technologies or solutions designed to reveal the origins of content, regardless of whether information has been attached to it. The deepfake detection market is still in its nascent stages, with increasing adoption across sectors despite significant barriers to market entry, adoption and scaling.

Demand. Deepfake detection is becoming critical across sectors for various reasons. Businesses are increasingly integrating deepfake detection for fraud prevention, brand protection, identity verification and content moderation. Governments and public authorities are leveraging these technologies to enhance national security, combat misinformation and disinformation, and support law enforcement investigations and prosecutions.

Supply. Since 2017, the global deepfake detection market has experienced rapid growth, led by 23 US-headquartered providers out of 59 identified third-party firms as of 2025, followed by 7 from the UK, with the remainder spread across other regions. Many of these providers remain in the pre-seed or seed stages of funding, with an average total funding of £25 million. The primary technical approaches focus on machine learning, particularly neural network architectures and feature-based detection methods.

Barriers. Key barriers to entry and adoption include:

Drivers. The future market direction will depend on the degree to which the following factors are present:

In 2025, the Department for Science, Innovation and Technology (DSIT) commissioned PUBLIC to investigate the state of the UK deepfake detection market and its potential.

Through a literature review of over 80 sources, 14 expert interviews, a workshop, and the mapping of 59 deepfake detection providers in the UK and globally, this report aims to develop a robust evidence base and inform how the government can support the continued growth of this market.

From December 2024 to March 2025, the Department for Science, Innovation and Technology (DSIT) commissioned PUBLIC, a specialist digital advisory firm, to undertake deep-dive research into the current state of the UK deepfake detection market to inform how the government might incentivise and grow this market.

Deepfake detection technology is rapidly evolving to combat threats posed by AI-generated content. While progress is being made through government initiatives, industry collaboration, technological solutions, and ongoing research, the effectiveness of detection tools remains a challenge due to the rapid evolution of deepfake technology.

Develop a robust evidence base on the UK deepfake detection market, analysing the market’s maturity, trends and the UK’s global positioning in deepfake detection.

Identify barriers to adoption and scaling of deepfake detection solutions and key factors driving future growth.

Explore the future evolution of the market and HMG/DSIT’s roles to support continued growth of the UK deepfake detection market.

While there is no universally accepted definition of deepfake, this report adapts Ofcom’s definition to include unintended deepfakes.

Video, image, text or audio that has been generated in whole or partly by Artificial Intelligence (AI) algorithms.

Deepfake is a subset of synthetic media. Audio-visual content that has been generated or manipulated using AI, and that misrepresents someone or something.

Created using AI , rather than simpler methods

Audio-visual content , rather than text

Misrepresenting someone or something Deepfakes may depict real or fictitious people, events, or objects

Potential to cause harm. Deepfakes can have the potential to cause harm, regardless of intent

(1): DRCF, The Future of Synthetic Media , 2024.

(2): Ofcom, Deepfake Defences: Mitigating the Harms of Deceptive Deepfakes , 2024. We adapted the definitions of synthetic media and deepfakes based on Ofcom’s definition.

In particular, this report focuses on deepfake detection.

What is the current state of maturity of the deepfake detection market in the UK, and how has it evolved with AI capability development? How is it expected to evolve in the face of international competition? Analysing the supply and demand in the UK deepfake detection market, including its composition, segmentation, growth, primary use cases in both public and private sectors, to assess the maturity of detection technology and the market.

What accuracy metrics are currently used to evaluate deepfake detection tools? What is the most sensible approach to measuring accuracy? Collating various accuracy metrics and key considerations for evaluation, to support the understanding of the efficacy of detection tools and the potential role of the government moving forward.

What are the key barriers and drivers to the commercial success for third-party deepfake detection providers? Performing a PEST analysis to legislative and regulatory, commercial, social and technological factors that are either slowing down or driving market entry and deployment.

How does deepfake detection technology interface with other aspects of the safety tech industry? Contextualising the deepfake detection market within the broader safety tech ecosystem to highlight the evolution of the trust and safety landscape.

How can the Government support continued growth of the UK deepfake detection market? Identifying the role the government can play in addressing existing barriers between deepfake detection demand and supply, and in preparing for the future.

We conducted a comprehensive review of diverse sources on the deepfake detection market landscape.

From the evidence base, we identified key market trends, barriers and drivers, using the PEST framework.

We then identified the best possible future for the market, and the pathway from its current state to that vision.

Finally, we developed recommendations on where government action or further work by DSIT would be impactful.

The deepfake detection market is still in its early stages, with increasing adoption in key use cases such as fraud prevention, identity verification, content moderation, national security, and law enforcement.

Despite notable growth since 2017, led by US and UK players, most providers remain in early-stage funding rounds and struggle to scale.

Key barriers to entry and adoption include regulatory uncertainty, high costs of technological development, resource constraints, low perceived return on investment (ROI), limited confidence in detection accuracy, and lack of training data.

The deepfake detection market is expected to continue growing and diversifying in terms of use cases over the few years, driven by several factors outlined in the section, although the pace may slow.

Notes: (1): The indicative market growth trend (in terms of the number of providers founded) is based on the analysis of 59 providers identified in this research.

Below are the key considerations for each of the seven primary use cases

Government, law enforcement, defence, and healthcare agencies are increasingly investing in detection tools to verify evidence, combat fraud, and misinformation and threats that harm public safety and democratic integrity

Public sector bodies, regulatory agencies, and electoral integrity bodies require deepfake detection to combat fraud, electoral manipulation, voice impersonation, and the interference of public service delivery. These tools are critical for safeguarding democratic processes, maintaining public trust, and ensuring the integrity of civic functions.

This sector requires deepfake detection to counter mis/ disinformation, system infiltration, intelligence deception, and terrorist threats , to maintain national stability. National security agencies, military bodies, and intelligence services rely on these tools to identify synthetic media that manipulates public perception, compromises critical infrastructure, and disrupts operations.

Law enforcement agencies require deepfake detection to combat deepfake-enabled crimes, including fraud, blackmail, and child exploitation, as well as to identify fraudulent activities and prevent evidence manipulation. These tools are essential for maintaining the integrity of investigations, ensuring fair legal processes, and protecting public safety from deepfake-enabled deception.

Healthcare providers and public health agencies need deepfake detection to prevent medical fraud, data breaches, and patient deception. Deepfakes can manipulate diagnostics, impersonate doctors, and spread false medical advice, endangering patient safety. Detection tools safeguard telemedicine, protect sensitive data, and maintain trust in medical communications.

From finance to media, private sector companies are adopting detection tools to prevent scams, financial losses, and reputational damage while countering synthetic media threats that erode security and consumer confidence

BFSI institutions require deepfake detection to combat sophisticated fraud, identity theft, voice cloning, and synthetic identity fraud. These tools are essential for securing authentication, detecting manipulated claims, preventing financial scams, safeguarding digital transactions, and ensuring trust in financial services and risk management processes.

Enterprises and risk management teams need deepfake detection to prevent fraud, insider threats, and corporate espionage , while protecting against reputational damage . These tools verify identities, safeguard sensitive communications, and maintain operational integrity in the face of manipulated media that could enable scams, data breaches, or reputational damage.

News, media agencies, and entertainment companies need deepfake detection to combat manipulated content, protect their reputation and brand, and counter misinformation and IP misuse. These tools are crucial for verifying content authenticity, preventing digital deception, safeguarding creative assets, and maintaining audience trust.

Social media platforms and online communities need deepfake detection to combat harmful content, misinformation, and harassment. These tools enforce policies, detect manipulated media, and protect users from deception and digital harm. Automated detection is crucial to maintaining trust and platform integrity at scale.

Note The initial scope of this project focuses on deepfake detection. Since deepfake detection tools are often paired with embedding tools and many providers develop both solutions, both categories are included in the following supply-side analysis.

(1): DSIT, The UK Safety Tech Sector: 2024 Analysis , 2024.

(2), (3): Ofcom, Deepfake Defences: Mitigating the Harms of Deceptive Deepfakes , 2024. Definitions adapted from those provided.

Tech or solutions to facilitate safer online experiences, and protect users from harmful content, or conduct.

Tech or solutions that aim to determine whether content was generated, manipulated, or altered using AI.

Tech or solutions designed to attach and provide contextual information to content at the point of creation or editing

Tech or solutions designed to reveal the origins of content, regardless of whether information has been attached to it.

Overall: The global deepfake detection market is experiencing rapid growth with the number of providers increasing nearly 380% since 2017 and US-headquartered firms leading the sector. Of the 59 third-party deepfake detection providers identified worldwide, 7 are based in the UK , positioning it as the second-largest economy in this market.

Growth: While the market began emerging in the early 2000s, dedicated deepfake detection providers only began to surface in 2017. Since then, the market has expanded by nearly 380% , largely driven by companies specialising in machine learning techniques for deepfake detection.

Size: The majority of providers (83.0%) are micro or small enterprises. Dedicated providers tend to be even smaller in terms of employee size and funding compared to diversified counterparts. This is mainly due to their relative youth and their progression through earlier stages of development.

Investment: Many of these providers remain in the early stages of funding , with an average total funding of £25 million. Specifically, 8 (36.4%) dedicated providers are in the pre-seed or seed stages, while only one (Reality Defender) is in Series A.

Segments: The providers mapped predominantly focus on fraud prevention (72.9%) and identity and age verification (45.8%) services to the financial industry, and mis/disinformation detection (50.8%) for government, media, or social media platforms.

Source: Data on this slide are based on PUBLIC’s analysis of 59 third-party deepfake detection providers using open-source research. See Appendix A2 for methodology details.

Specialist Sole Focus on Deepfake Detection These providers dedicate themselves exclusively to the development of deepfake detection technologies, services and solutions, establishing themselves as specialists in this field. Examples: Reality Defender GetReal Labs

Expansion Building on Adjacent Capabilities Some providers, particularly those already established in adjacent fields, are expanding their service offerings to include deepfake detection. Through internal R&D and strategic acquisitions , these providers naturally extend their capabilities to meet the growing demand from clients seeking deepfake detection solutions. Examples: SURF VIDENTIFIER

Partnership Leveraging External Capabilities Rather than develop deepfake detection technology in-house, some providers partner with specialists in this field, given internal resource constraints. This allows them to integrate advanced deepfake detection tools into their existing services, broadening their offerings without the need for significant technology development investment. Examples: Proof (Partnership with Reality Defender)

Partnership with online safety technology Deepfake detection providers partner with other deepfake detectors or online safety tech companies to offer comprehensive solutions for detecting and preventing AI-generated threats, combining specialised expertise in deepfake detection with content moderation, identity verification and fraud prevention capabilities. For example: DeepTrust and Breacher.ai announced a partnership to combat deepfake AI attacks, offering a joint solution, DeepBreach, for forensic analysis of potentially fraudulent content. (1) iProov’s liveness detection technology is integrated into Jumio’s identity verification and KYC platform to combat sophisticated cyber attack using deepfakes. (2)

Partnership with AI developers & GenAI services Deepfake detection providers partner with AI developer and GenAI services to improve their detection algorithms and accuracy, but also to stay ahead of evolving threats. For example: Reality Defender partners with ElevenLabs , an AI voice generator, to advance audio deepfake detection models, and leverage ElevenLabs’ proprietary technology to enhance its detection tools. (3) DeepTrust partners with Musicify , a voice clone and AI music generation service, to join forces on responsible AI usage and development, focusing on AI music detection and IP protection. (4)

(1): DeepTrust, DeepTrust and Breacher AI Partnership Announcement , 2024.

(2): Jumio, Jumio Adds iProov’s Award-Winning Liveness Detection to its KYX Platform , 2021.

(3): Reality Defender, Strengthening Digital Trust: How ElevenLabs Helps Reality Defender Advance Voice Deepfake Detection , 2025.

(4): DeepTrust, DeepTrust and Musicfy Partnership Announcement , 2024.

Tech giants have developed their own deepfake detection (and identification) solutions, with examples such as:

Detection Microsoft Video Authenticator (1)

Microsoft Video Authenticator can analyse a still image or video in real time to provide a confidence score that the media is artificially manipulated. It uses machine learning algorithms to detect subtleties like pixel inconsistencies and colour fading.

The tool, launched in 2020, is developed In partnership with Reality Defender to combat AI-generated disinformation.

Detection Intel’s FakeCatcher (2)

Intel’s FakeCatcher is a real-time deepfake detector that analyses biological signals to determine the authenticity of a video. FakeCatcher uses Photoplethysmography (PPG) - a technique that analyses ‘blood flow’ in video pixels to determine a video’s authenticity in milliseconds.

The tool was first launched in 2022.

Identification Google Deepmind’s SynthID (3)

SynthID watermarks and identifies AI-generated content across various media types, including images, audio, text and video , by embedding digital watermarks directly into the content during generation and using deep learning models for subsequent identification.

The Beta version (current version) was launched in 2023 and was made available in open source.

Identification Meta Video Seal (4)

Meta Video Seal is a neural watermarking tool that embeds durable, invisible watermarks for AI-generated videos - even after video editing. The tool also allows for embedding up to a 6-character hidden message within the watermarks.

Video Seal was launched in December 2024, following the release of Audio Seal in June. Video Seal is available in open source.

(1): Microsoft, New Steps to Combat Disinformation , 2020.

(2): Intel, Real-time FakeCatcher for Deepfake Detection .

(3): Google Deepmind, SynthID .

(4): Meta, Video Seal - Meta FAIR AI Demos

Regulatory uncertainties, delays, and compliance barriers challenge third-party providers despite global legislative efforts.

Regulatory lag and ambiguity UK market entry and compliance hurdle

High technical costs, resource constraints, and low ROI perception of T&S remain big barriers to deepfake detection innovation.

Customers perceiving low ROI in deepfake detection High costs and resource constraints

There is rising public awareness of deepfake threats, but scepticism toward detection tools persists.

End user scepticism toward detection tools

GenAI and ML advancements drive detection solutions, but limited datasets, inconsistent metrics, and evolving threats remain challenges.

Low access to high-quality training datasets Heterogeneity of testing datasets and metrics

Political deepfake threats eroding institutional trust: Deepfakes diminish institutional trust by enabling the spread of false and damaging mis/disinformation, a concern heightened by 2024’s global election cycle. This growing threat drives political and regulatory action to mitigate its risks. (1)(10)

Increasing global efforts on legislation against deepfakes: Governments worldwide are ramping up efforts to regulate deepfakes. In the UK, the upcoming Crime and Policing Bill would ban AI-models optimised to produce CSAM. The EU is enforcing transparency requirements through the AI Act and Digital Services Act. In the US, states like California are enacting targeted laws, alongside federal efforts like the NO FAKES Act and Take It Down Act. Spain plans to fine unlabelled AI-generated content and criminalise AI-generated sexual images. Countries like Australia and South Korea have also implemented similar measures. (2)(3)(4)(5)(6)(7)(8)(12) These are driving increased compliance demands and regulatory scrutiny.

Regulatory ambiguity is leaving deepfakes a low priority for online platforms: Despite growing regulatory momentum, ambiguity remains. While the Online Safety Act (OSA) addresses illegal harm, including both deepfakes and non-deepfakes, the safety tech industry perceives a lack of clarity around its application to deepfakes or synthetic media. This perceived ambiguity prompts platforms to take a reactive, ‘wait and see’ approach to addressing deepfakes. (7)(12)

Regulations lagging behind deepfake threats and technology: AI advancements, including deepfake technology, normally evolve faster than regulatory frameworks can adapt. As a result, regulatory responses currently remain limited, creating uncertainty and reluctance for businesses and providers to act. (8)(9)(12)

UK market entry and compliance hurdles faced by foreign players: As noted by a few US-based interviewees, market entry and compliance obstacles, coupled with complex public procurement processes, create challenges for foreign providers trying to access the UK market. (12)

(1): Kharvi, Understanding the Impact of AI-Generated Deepfakes on Public Opinion, Political Discourse, and Personal Security in Social Media , 2024;

(2): Herbert Smith Freehills, Criminalising deepfakes – the UK’s new offences following the Online Safety Act , 2024;

(3): Rouse, AI-generated deepfakes: what does the law say? , 2024;

(4): Morrison Foerster, 2024 Year in Review: Navigating California’s Landmark Deepfake Legislation , 2024;

(5): Reed Smith, AI and publicity rights: The No Fakes Act strikes a chord , 2024;

(6): Live Now Fox, What is the Take It Down Act? , 2025;

(7): Reuters, Spain to impose massive fines for not labelling AI-generated content , 2025;

(8): Tech Xplore, Spain seeks to criminalize AI-generated sexual images , 2025;

(9): Labuz. M., Deep fakes and the Artificial Intelligence Act—An important signal or a missed opportunity? , 2024;

(10): Howard Kennedy, The Regulatory Gap in Deepfake Technology , 2025;

(11): GDPR Local, Deepfakes and the Future of AI Legislation: Overcoming the Ethical and Legal Challenges , 2025.

(12): Interviews, 2025.

Rising enterprise losses calling for deepfake fraud prevention: Deepfakes are escalating financial risks across industries, enabling harms such as fraud, IP theft, brand damage, and executive impersonation, leading to significant economic losses for enterprises. (7) The inclusion of deepfake-enabled fraud is projected to drive total fraud losses in the U.S. to $40 billion by 2027, up from estimates that previously excluded AI-related threats. (1)

Provider willingness to partner: Industry players are increasingly forming partnerships with private and public entities to develop more effective solutions for addressing the diverse harms posed by deepfakes. (2)(3)(4)(5)(6) The UK Government is collaborating with industry and researchers to advance deepfake detection, supporting initiatives like the Deepfake Detection Challenge to address technological and regulatory challenges. (6)

Government funding on frontier AI safety driving foundational research: Government funding for frontier AI, such as the AI Security Institute’s Systemic AI Safety Grants, supports research addressing AI-related risks, including emerging threats like deepfakes. (7)

Customers perceiving low ROI in deepfake detection: For customers, deepfake detection competes for resources against higher-priority areas such as product development, growth, and operational efficiency. As a result, investment in trust and safety often remains limited, slowing adoption and innovation in deepfake detection solutions. (7)

High costs and resource constraints: Keeping pace with deepfake threats demands ongoing investment in AI research, computing power, and expertise. High development costs pose a barrier, especially for smaller deepfake detection providers, limiting market entry and innovation. (7)

(1): Deloitte Center for Financial Services, Generative AI is expected to magnify the risk of deepfakes and other fraud in banking , 2024;

(2): Reality Defender, Partners , 2025;

(3): Continuity Insights, New Partnership Will Advance Audio Deepfake Detection , 2024;

(4): PR Newswire, Reality Defender Partners with TaskUs to Expand Deepfake Detection to Content Moderation Teams and Call Centers , 2024;

(5): Hive, Announcing Hive’s Partnership with the Defense Innovation Unit , 2024;

(6): TechUK, UK Launches £200,000 Grants for Systemic AI Safety: Fostering Trust and Safety , 2024;

(7): Interviews, 2025.

Increased public awareness of deepfake threats: There is rising public awareness and concern deepfakes, particularly online CSEA, driven by high-profile cases and media coverage, (1)(2) potentially pushing political and policy attention toward addressing deepfakes.

Reputation: End user trust is a strong demand incentive for deepfake detection technology across a wide range of sectors. As new GenAI technology increases the number of crimes in various industries, both public and private institutions are more inclined to invest in deepfake detection services to maintain trust in their reputation. (4)

(1): The Alan Turing Institute, Behind the Deepfake: 8% Create; 90% Concerned , 2024;

(2): Kharvi, Understanding the Impact of AI-Generated Deepfakes on Public Opinion, Political Discourse, and Personal Security in Social Media , 2024;

(3): Business Wire, iProov Study Reveals Deepfake Blindspot: Only 0.1% of People Can Accurately Detect AI-Generated Deepfakes , 2025;

(4): Interviews, 2025.

Increased GenAI-enabled threats: The rapid advancement of generative AI (GenAI) models - combined with their increasing realism, scalability, and accessibility - is accelerating deepfake-related threats, pushing the need for continuous innovation in detection technologies to keep pace. (1)(2)(3)(4)(5)(6)(10)

Development in AI advancing machine learning (ML) based deepfake detection: Ongoing innovations in AI-driven detection techniques are enhancing the ability to identify and combat deepfakes. Improved machine learning models, multimodal detection approaches, and real-time analysis capabilities are driving more effective and scalable solutions, strengthening the deepfake detection market. (7)(10)

Limited access to high-quality training datasets: Most detection tools are trained on datasets that often lack real-world synthetic media, have limited coverage, and feature a disproportionate representation of deepfakes. Testing on these datasets can lead to inaccurate results. As noted by one interviewee, accuracy rates typically drop by 10-20% in real-world redeployment with representative datasets, compared to lab environments using training datasets. (8)(10)

Heterogeneity of testing datasets and metrics: Inconsistencies in testing datasets and evaluation metrics result in low confidence in the efficacy of detection tools, while providers often claim high accuracy without independent validation. (9)(10)

(1): techUK, Detecting deepfakes: A roadmap to UK resilience in the face of GenAI , 2024;

(2): Tech Target, How to prevent deepfakes in the era of generative AI , 2024;

(3): Visua, The Deepfake Detection Arms Race ;

(4): DRCF, The Future of Synthetic Media , 2024;

(5): Deepmedia, How Deepfakes Will Challenge Biometric Face Verification in 2025 , 2025;

(6): Biometric Update, Deepfake ecosystem develops around apps, services as detection fights to keep pace , 2024;

(7): GAO, Combating deepfakes , 2024,

(8): Yan et al., DeepfakeBench: A Comprehensive Benchmark of Deepfake Detection , 2023;

(9): UK Government, Innovating to detect deepfakes and protect the public , 2025;

(10): Interviews, 2025.

In this research, drivers are defined as critical uncertainties that could impact the deepfake detection market in the future.

Nine key drivers were identified, which interact to shape both demand and supply in the market. These include:

● Rapid advancement and availability of GenAI ● Threat to national security ● Regulation and enforcement mechanisms ● Customer understanding of deepfakes ● Changing user behaviours in response to deepfakes ● Market entry conditions for foreign vendors ● Available R&D investments ● Access to high-quality training data ● Standardised accuracy testing

Synthesised from desk research of 86 sources and 14 expert interviews, primarily with deepfake detection providers

As deepfake technology advances and related crimes become more frequent and financially significant, consumer awareness - among individuals and companies alike - has grown over the past few years.

This shift has altered their cost-benefit estimations, driving greater demand for deepfake detection tools and services.

Further GenAI advancement would escalate threats and risks, likely driving up demand for deepfake detection technology, incentivising investment , and ultimately improving the quality of deepfake detection.

Deepfakes hold the potential to diminish trust in public institutions and influence public opinion and elections by generating misleading content on sensitive topics and prominent figures. Public authorities are making early investments in detection technologies (e.g., via grant and challenge funds) and considering regulatory measures.

Governments worldwide are increasingly aware of the political risks of deepfakes . This phenomenon is increasing the deepfake detection technology demand among authorities globally.

Existing legislation, such as the EU AI Act, the UK’s OSA, and various state-level regulations in the US, have laid the foundations in regulating against harmful content. If implemented effectively, these measures are likely to drive higher demand for detection technologies.

Stronger regulations could shift the ROI calculation for deepfake detection customers due to the fear of fines, therefore driving demand across sectors from finance to social platforms, especially where end users are more exposed to harmful synthetic content.

Not every individual or industry exposed to deepfakes (whether through misinformation, disinformation, fraudulent intent, or public/private cyber theft) is aware of their multiple threats, nor are they familiar with detection tools. Increasing awareness of the potential risks posed by this type of content could drive greater demand for deepfake detection technologies.

As deepfake-enabled crimes become more frequent, consumers’ understanding of deepfake threats is likely to increase, driving up the demand for both preventive and detection tools , and boosting investment in technology.

Currently, deepfakes can target a variety of devices and scenarios. For example, criminals can reach individuals via mobile phones or organisations through hiring processes. This has broadened the use cases for deepfake detection and increased the need for novel detection techniques.

The rise of deepfakes in professional and personal contexts would change how individuals interact with technology, making them more cautious of potential deepfake threats . This is likely to drive greater demand for detection solutions, putting pressure on businesses and governments to implement change.

UK market but face compliance hurdles

Public procurement processes in the UK can be challenging for foreign providers to navigate. For example, a few US providers have reported feeling restricted from participating in challenges and procurement processes compared to their UK counterparts as it requires them to join UK government frameworks.

There also other collaboration challenges due to data sharing and protection as US providers do not have GDPR.

Positioning the UK as an attractive market for foreign vendors by simplifying deepfake detection procurement and providing support would enhance competition, promote knowledge sharing , and encourage innovation .

Public and private investment in new detection technologies can enhance the supply of deepfake detection services, while simultaneously increasing consumer confidence in these tools, potentially boosting their demand.

Resource availability can limit economies of scale in AI-based technologies, as they often need significant computing power, resources, and specialised knowledge. More R&D investment in deepfake detection would help overcome these resource constraints and drive supply (both in quantity and diversity).

Moreover, government-led R&D investment can signal market potential, providing greater certainty to both providers and private investors, encouraging market entry.

From the public investment perspective, consumption of deepfake detection solutions would increase as public investment in R&D improves access to this technology.

Government signalling could spark interest from the private sector, driving demand and investment in this technology. This would not only boost confidence but also generate evidence and research on its performance.

The increase in R&D on deepfake detection would boost the supply and confidence of the deepfake detection technology.

detection technology, building trust and increasing demand

The lack of updated, representative datasets creates two main issues. First, it prevents deepfake detection tools from training on data generated with the latest creation technologies, making them incapable of detecting new nuances. Second, it undermines the reliability of testing results against such datasets, resulting in low trust and confidence in the detection tools.

The availability of better training datasets can influence the deepfake detection market by enhancing the accuracy and effectiveness of detection tools. High-quality, diverse, and up-to-date datasets allow detection providers to develop more robust algorithms, improving the reliability of their solutions across a wider range of deepfake types and use cases.

This, in turn, increases the trust of both consumers and businesses in these technologies, driving demand.

The availability of high-quality training data is crucial for expanding the supply of deepfake detection solutions and improving their quality through better training. As a result, consumer trust would increase , driving higher demand.

There is no consensus on the accuracy metrics that providers use to promote their services, leading to consumer uncertainty and reluctance to pay for services they do not fully trust. Establishing a standardised way to evaluate accuracy to harmonise product results would enhance consumer trust and, in turn, drive increased spending on this type of technology.

The lack of clarity on accuracy metrics for deepfake solutions has led to consumer scepticism. Mitigating doubts effectiveness, including the use of different datasets and metrics, is key to building trust.

The nuance of real-world applications means that accuracy metrics will meaningfully differ depending on the use case in which deepfake detection tools are applied.

A standardised approach and guidance on accuracy evaluation would benefit both consumers and providers . It would align key considerations when conducting technical testing and interpreting the results, enabling meaningful comparison among different tools.

Barriers: Obstacles to entry and adoption in the deepfake detection market.

Deepfakes: Audio-visual content that has been generated or manipulated, and that misrepresents someone or something. (1)

Detection: Measures to identify and detect deepfake content after they have been created and shared. (1)

Drivers: Factors or influences shaping the future of the deepfake detection market. Drivers collectively impact the demand and supply of the detection market.

Embedding and Identification: Measures to attach and provide contextual information to content at the point of deepfake creation or editing to enable distinguishing them from real media. (1)

AI: Artificial Intelligence

CSEA: Child Sexual Exploitation and Abuse

LLM: Large Language Model

OSA: Online Safety Act 2023

ROI: Return on investment

Sources: (1): Ofcom, Deepfake Defences: Mitigating the Harms of Deceptive Deepfakes, 2024

Reviewed 86 relevant academic, industry, government and civil society reports, policy papers, and journal articles.

Conducted 14 one-to-one, semi-structured interviews , subsequently compiling the insights into key barriers and drivers in the deepfake detection market. Targeted a mixed set of safety tech providers prioritised by geography and organisational size.

Identified challenges and drivers in the deepfake detection industry.

Key findings highlighted the components and motivations on the demand side, as well as pain points on the supply side. These insights informed this report.

Developed a futures workshop with 13 participants from government, regulators, and deepfake detection providers to validate drivers, define the current and desired deepfake detection market scenarios.

Below is a list of organisations engaged to gather insights through the course of this research:

The workshop was not predicting a certain future. It was exploring a range of plausible futures, strengthening future facing policies and strategies, and encouraging collaboration and trust among public sector stakeholders.

A list of 9 drivers affecting the future was identified based on desk research and expert interviews.

Which drivers should we pay attention to? Which drivers can the government feasibly influence?

What are the main players in this scenario? How would each key driver play out in the scenarios? Is it a scenario we want or not?

Several limitations should be noted alongside the research findings

The provider mapping built upon research into the safety tech ecosystem, including the DSIT’s UK Safety Tech Sectoral Analysis. Additional providers were identified via available web data, online aggregators (e.g., Crunchbase), and manual keyword searches and reviews.

The focus of this project was on qualitative analysis over quantitative analysis, resulting in gaps in the economic data and the quantification of the size, growth and benefits of deepfake detection.

The provider mapping aimed to establish a foundational baseline for the sector’s future development. As such, it should not be viewed a one-time effort or an exhaustive list of providers.

The mapping focused on third-party providers, while recognising that several larger tech companies like Microsoft, Intel, Meta, and Google DeepMind develop in-house detection solutions (as shown in slide 25).

Stakeholder engagement:

The stakeholder engagement for this research primarily focused on expert interviews with industry providers who have established deepfake detection capabilities or are currently exploring these capabilities, focusing on the supply side of the market.

Engagement with the demand side was relatively limited , with a small group from the public sector (e.g., DSIT, NCA, Ofcom) participating in a workshop or interviews. As a result, there is a gap in the direct engagement with other key stakeholders and customers of deepfake detection from both government and industry sectors (including GenAI developers and application providers). This presents potential for future engagement.

This work was prepared by a specialist digital advisory firm, PUBLIC, who have been supporting the department with this analysis. PUBLIC have designed and delivered a number of UK Government programmes designed to support high-growth technology sectors, including with the Department for Science, Innovation and Technology, the Ministry of Housing, Communities and Local Government, the Ministry of Justice, the NHS, and the Geospatial Commission.

This work is not official government policy.