veraAI concludes three-year project; verification tools now available for newsrooms

veraAI wraps: production-ready verification tools for your newsroom and engineering teams

After three years of R&D on AI-based fact-checking tools with media professionals, the Horizon Europe project veraAI is concluding. The output is clear: practical tools you can use now to assess text, audio, images, and video, plus methods and training to support responsible deployment.

The EBU, together with Deutsche Welle, contributed across the full lifecycle: participatory design with users, algorithm development, iterative evaluations, and hands-on dissemination through webinars, workshops, and demos.

What the collaboration delivered

• Participatory design workshops to surface user needs and translate them into requirements, then validate through structured evaluations.
• Algorithms for audience and authorship estimation.
• Ongoing communication through channels including the tech-i app, plus a webinar series on the research behind the tools.
• Methodology and communication training for the consortium.
• Training workshops for Members and a catalogue of video tutorials.
• A report on the potential of C2PA and veraAI tools.
• A legal and ethical handbook for developers and deployers of AI-based fact-checking systems.
• Presentations and demos at events like AI Summit, DTS, PTS, and the EBU zone at IBC.

Get the veraAI tools for your organization

Access the tools via two primary platforms, with a few standalones currently available and more integrations on the way.

• Verification Plugin (VP): A free Chrome extension with verification utilities for text, audio, and video. Built for journalists, fact-checkers, and human rights defenders, and open to the public. Install it via the Chrome Web Store by searching "Verification Plugin (veraAI)".
• Truly Media (TM): A collaborative verification platform to collect web content, organize evidence, and run analysis through integrated tools. Fill in the platform's interest form to start the registration process or request a demo.

What's available now

Below is the current lineup. VP = Verification Plugin, TM = Truly Media, standalone = separate service.

Synthetic media detectors

• Synthetic Image Detector: Likelihood that an image is AI-generated (VP + TM).
• Synthetic Audio Detector: Likelihood that parts of an audio file contain synthetic speech (TM).
• Machine-Generated Text (MGT) Detector: Probability reports for segments of text (VP "Assistant"), plus an overall score (VP + TM).
• Geolocalizer: Visual clue extraction from an image to estimate its location (VP).
• Image Forensic AI Filters: Indicators of tampering in manipulated images (VP).
• Deepfake Video Detector: Probability that a video contains AI-manipulated faces (swap or reenactment) (VP).

Analytical support

• Database of Known Fakes: Searchable corpus of already reviewed claims from trusted fact-checking and news organizations (standalone).
• Credibility Signals: Text analysis that flags persuasion techniques with confidence levels (VP "Assistant", alongside MGT).
• Coordinated Sharing Detection Service: Network analysis of coordinated sharing patterns on social platforms (standalone).
• Keyframe Selection and Enhancement Service: Extracts keyframes and enhances faces/text to speed up video review (standalone + TM).

Implementation notes for IT and development teams

• Browser rollout: VP is a Chrome extension-plan for extension policy, permissions, and version pinning. Test with managed profiles before fleet deployment.
• Data handling: Align tool use with your data protection policies. Confirm whether processing runs locally or via remote services and document retention settings.
• Scoring interpretation: Outputs are likelihoods, not ground truth. Combine multiple signals (e.g., MGT + Credibility Signals + Known Fakes) and document thresholds by use case.
• Workflow integration: For collaborative investigations, use TM to centralize evidence and tool outputs. For high-volume video, pair Keyframe Selection with Deepfake detection to cut review time.
• Evaluation: Establish a baseline set of known samples and track precision/recall by content type. Re-test after updates to models or plugins.
• Governance: Reference the project's legal and ethical handbook to set internal guidance for deployment, access control, and audit trails.

Standards and further reading

The project includes analysis of open provenance standards that help authenticate media across the chain of custody. Explore the C2PA specification here: c2pa.org.

For background on the funding program, see Horizon Europe.

Availability, updates, and support

An up-to-date tool list will be posted on the project website next month. To discuss access for your team or clarify technical details, contact Hans Hoffmann (EBU Technology & Innovation) and/or the veraAI project lead. Training videos and webinar recordings are available to help your team get productive fast.

Funding

The veraAI project is co-financed by the European Union, Horizon Europe programme, Grant Agreement No 101070093, with additional funding from Innovate UK grant No 10039055 and the Swiss State Secretariat for Education, Research and Innovation (SERI) under contract No 22.00245.

14 Sep 2025