Veraquo: integrating content verification into everyday decision-making

Juan González Martínez, Director of Security and Privacy at Gradiant, explains how a need already emerging in newsrooms and organisations led to the development of an intelligent digital content verification platform.
Consider this scenario: a newsroom receives a video and, before sharing it, must assess whether the footage is genuine or has been manipulated for a particular purpose. Or take an organisation reviewing a document on which the approval of an application depends. Although the people making these decisions work in different fields, they face the same question: what can they verify about the file before taking action?
At Gradiant, we developed Veraquo to bring this verification capability into the processes where it can make a difference. The platform analyses images, videos, audio and documents for signs of manipulation or generation using artificial intelligence. Its development began with a question that guides our applied research: how can we turn a technical capability into a tool people can use when they need it?
For non-specialists: How does Veraquo detect a fake video or voice? What details, invisible to the human eye, does the tool look for?
The key to Veraquo is that it analyses an image, video or voice differently from the way people do. When we watch a video or listen to a voice, we consider whether a face looks familiar, a voice sounds natural or a scene seems believable. Veraquo, by contrast, can identify digital patterns and inconsistencies that are completely imperceptible to our senses.
Most importantly, a forgery does not have to be obvious for us to suspect it. This is precisely why we develop these technologies at Gradiant: fake content is becoming increasingly realistic, and we can no longer rely solely on what we see or hear.
Juan, what market need did you identify that led you to develop Veraquo?
We saw that more and more decisions depended on material received digitally by organisations. A video could become a news story, an audio recording could support an urgent request, and a document could allow an administrative process to move forward. The person making the decision needed a way to examine that material more closely, within the time and resources available.
That presented an opportunity for Gradiant. We had been researching artificial intelligence and digital forensics for some time, and we could apply that knowledge to a tool that fitted into teams’ workflows. We asked ourselves a very practical question: what information can we give someone looking at a file so that they can make a better-informed decision about what to do with it?
And what technical decisions did you make in response to that need?
One was to bring the analysis of different content types together on a single platform. Take a video, for example: we may want to examine what appears on screen, but also the accompanying voice. Each format provides different signals. Combining these capabilities means that the team receiving the material can analyse it without having to find a separate tool for each file.
Another decision was to design the technology on the understanding that generative models would continue to evolve. When we analyse a file, we look for signs that may go unnoticed when watching or listening to it. To keep that capability useful as new models emerge, we are working on an approach that can adapt without having to retrain the system on large volumes of data every time. This adaptability was part of the challenge from the outset.
How would a newsroom use Veraquo after receiving a video?
Imagine a video of a current event arriving through a messaging app, with the team having little time to decide whether to include it in a news report. The analysis may flag signs that warrant spending more time examining the file or requesting the original version. The newsroom can compare that result with other checks: who recorded the video, when it was recorded and whether what it shows matches information from other sources.
In that context, the value lies in providing information in time to guide the verification process. The editorial decision remains with the team, which can assess the file analysis alongside its provenance and the context in which it was received.
And what changes when we talk about intelligent verification in a business?
What changes is the decision that needs to be made. Consider a team receiving documentation to register a new customer or approve a loan application. If the analysis flags a possible alteration, that result can trigger an additional review of the document. Depending on the company’s procedure, the team might cross-check it against another source or request a copy through a different channel.
That example also illustrates the value of integration. If an organisation analyses files occasionally, it can use the web interface. If it receives documents regularly, it can integrate the analysis into its existing process through an API and decide when a person needs to intervene. Verification can then become part of the workflow from the moment a document arrives.
How should a team interpret the analysis results?
As an indication to help decide what to do next. If signs of manipulation appear, the team can examine the file more closely and cross-check it against the available information. And if the analysis finds no such signs, the file’s origin and context remain important, particularly when the decision has significant consequences.
That is why, when introducing a verification tool, it is important to consider what happens next. Who reviews an alert? What other information can they consult? When should a second check be requested? Two organisations may receive similar results but respond differently because the cost of getting it wrong differs. The analysis provides one piece of information that each team can use according to its responsibilities and procedures.
Generative AI is advancing rapidly. How accurate is Veraquo today, and how is it trained to keep pace with new synthetic content generation models?
This is one of the major technological challenges ahead of us. We are seeing constant change: increasingly convincing synthetic images, videos and voices are being produced, and this means we must continually improve detection systems.
With Veraquo, we addressed this challenge from the design stage, developing patent-pending technology that can adapt rapidly to new generative AI models and versions without needing to retrain the system using large volumes of data.
This is particularly important because new tools capable of generating fake content are constantly emerging, and detection systems need to evolve at the same pace.
Veraquo currently achieves high levels of accuracy in detecting AI-generated content, with results exceeding 90% in certain evaluations. Accuracy depends on the type of content and the analysis conditions. But beyond the current figures, what really matters is our ability to maintain these results as generation technologies evolve.
At what point does research become technology that makes an impact?
For us, it happens when that capability informs a specific decision. This also reflects Gradiant’s market-oriented approach. Understanding how organisations work helps us decide what to research, which capabilities to bring together in a product and how to make them available to those who need them. With Veraquo, this approach connects a line of research with decisions that affect the quality of the information we receive and the security of the services we use
What would need to happen for this verification capability to reach the general public as well?
The next step we envisage is bringing it closer to the point where doubts arise. Today, Veraquo is primarily aimed at professional use. In the medium term, capabilities like these could be integrated into messaging applications, social media or devices themselves. This would allow people to receive an alert as soon as suspicious content arrives, while there is still time to decide how to respond.
If someone receives an audio message that sounds like a family member urgently asking for money, what matters is having time to verify the request before making a transfer. An alert could give them the opportunity to call that person on a known number. Strengthening digital trust means giving people that ability to act, including when decisions are made on a mobile phone in a matter of minutes.
Veraquo won a 2026 CyberSecurity Breakthrough Award in the United States and was presented in San Francisco. What does it mean for Gradiant and Galicia’s technology ecosystem to be at the forefront of the global fight against deepfakes?
This recognition is an important endorsement of the work we have been doing at Gradiant for years. It shows that we can develop internationally competitive technology in Galicia in an area as demanding and strategically important as cybersecurity and artificial intelligence.
And it is not an isolated success. In 2024, we also took first place in one of the international deepfake detection competitions held as part of ACM Multimedia in Australia. The 2026 CyberSecurity Breakthrough Award now brings further recognition, this time for a market-oriented technological solution.
For us, both milestones reflect something especially important: our ability to turn high-level research into technology that addresses real problems faced by businesses and organisations.