Â鶹Éç

In an era where artificial intelligence is redefining how we see and understand the world, the Â鶹Éç Research & Development computer vision team is leading the charge. From pioneering deepfake detection to agentic AI, our mission is to ensure that audiences can continue to trust what they see, hear, and share.

The rise of generative AI has unlocked extraordinary creative possibilities, but it’s also made . Hyper-realistic 'deepfakes' can manipulate voices and images, blurring the line between reality and fabrication. For the Â鶹Éç - a global broadcaster whose reputation is built on accuracy and integrity - this challenge needs urgent attention.

That’s why the Â鶹Éç Research & Development computer vision team is pioneering AI-driven verification tools that help journalists, editors, and content producers authenticate media in real time. We’re building models that don’t just aim to detect manipulated content but to explain it, equipping journalists with the confidence to verify before they publish.

Every line of code we write, every dataset we build, and every prototype we release contributes to one goal: safeguarding trust in the digital age.

Deepfake detection

Our flagship deepfake detection pilot is setting new standards for how AI can serve journalism. We’re developing prototype tools that identify AI-generated or manipulated images and videos, giving journalists the ability to spot fabrications that might otherwise go unnoticed.

Unlike many third-party services, we build our models in-house. This gives us full transparency and control over data, algorithms, and outputs - a crucial advantage when working with sensitive or confidential content. It also lets us customise features for editorial needs – like explainability and trust indicators – and build them straight into newsroom workflows.

We collaborate closely with Â鶹Éç Verify, the organisation’s team dedicated to fact-checking and authenticity verification. Selected Verify journalists have been testing our prototype, helping us refine the tool with real-world feedback. We’re also working with Â鶹Éç Studios to evaluate how our detector can flag AI-generated content submitted by users before the Â鶹Éç shares or highlights them to audiences of our programmes or to our followers on social media.

And our innovation doesn’t stop at the newsroom. Our deepfake detection work has caught the eye of Weather Watchers, a Â鶹Éç community platform where users submit photos of local weather. By combining our AI detector with digital watermarking standards like C2PA, we can help editors instantly verify the authenticity of user-generated imagery, ensuring accuracy at every level of the Â鶹Éç.

The science behind the trust

We ground our approach to deepfake detection in rigorous science and collaboration. We’ve built the largest proprietary dataset of its kind, containing over one million examples of partially manipulated images. This data underpins our in-house models, which use foundation models to detect manipulation.

We’ve also conducted extensive benchmarking of existing detection tools to assess the performance, bias, and reliability of leading commercial systems. This commitment to transparency ensures that our technology is not just effective but also trustworthy and accountable.

NeurIPS 2025: Recognition on the global stage

Our research isn’t just making an impact within the Â鶹Éç. It’s earning recognition from the global AI community. In collaboration with the , we submitted and had our poster '' accepted at NeurIPS 2025, one of the world’s most prestigious AI conferences.

The work introduces RADAR, a groundbreaking method that leverages multi-modal signals to detect inpainted or diffusion-edited regions with unprecedented accuracy. This achievement highlights how industry-academic collaboration can drive meaningful progress in AI safety and media authenticity.

For our team, NeurIPS isn’t just a milestone, it’s a message. It shows that Â鶹Éç R&D’s AI research can stand shoulder to shoulder with the best in the world, and that the skills and creativity of our researchers are shaping international conversations about the responsible use of AI.

The next challenge

While our image detection tools are now being used across multiple Â鶹Éç divisions, our next frontier is video deepfake detection - a rapidly emerging area with immense social impact.

Conversations with Â鶹Éç Verify journalists revealed a pressing need for technology capable of verifying not just still images but moving footage. Our early experiments are already producing promising results. We’re now adapting our proprietary model to handle both image and video detection, exploring how temporal information can enhance overall performance.

As part of this evolution, we’re building a bespoke video test dataset designed to ensure fair evaluation against existing commercial models. Other companies have already absorbed many open-source datasets into their training pipelines, so we’re taking a more careful and transparent approach.

Collaboration

Our work thrives on collaboration. We maintain close relationships with leading universities, including the , , and , as well as organisations such as , the , and the .

These relationships keep us at the cutting-edge of AI research, while grounding our work in real-world applications. They put our models through rigorous testing and peer review to ensure that they meet the Â鶹Éç’s standards of accuracy and integrity.

The road ahead

At Â鶹Éç R&D, we won’t just build technology, we help redefine how the world tells truth from fiction.

We’re a team of researchers and creative technologists working together to use AI responsibly and for the public good. Our work sits at a rapidly developing place where technology and journalism meet. The challenges involved are complex, and .

As AI continues to evolve, so will we. In the coming months, we’ll be rolling out new versions of our detection tool prototypes, planning integration across Â鶹Éç platforms, and expanding our research into cross-modal detection where audio, image, and video insights come together for even more reliable verification.

We’ll continue to share our findings, publish at leading conferences, and engage the public in conversations about AI transparency because trust in media is not just a Â鶹Éç priority, it’s something the world depends on.