Research Programme

The central question of our programme: what constitutes proof that a digital image was captured from reality rather than generated? This theme develops the theoretical foundations and practical criteria for such proof.

We study cryptographic commitment schemes, sensor-level attestation, and the provability properties of different capture architectures. The output is a formal framework — a set of conditions that a capture system must satisfy for its outputs to be considered evidentially authentic.

This work is directly upstream of all other themes: without a clear definition of provable reality, detection, timestamping, and forensics are all weakly grounded.

Open Questions

• Can sensor hardware produce unforgeable attestation signals?
• What is the minimum cryptographic commitment required for legal evidentiary standards?
• How does provability degrade across format conversions and platform processing?

Blockchain-based timestamps are among the most robust mechanisms for establishing temporal precedence of a digital record. This theme evaluates whether NFT and broader blockchain timestamping mechanisms meet the evidentiary standards required in legal proceedings, insurance claims, and journalistic verification.

We analyse different chain architectures, smart contract patterns, and custody models, and assess each against the admissibility criteria of EU and international legal frameworks. We also study attack vectors — chain reorganisation, key compromise, and metadata spoofing — and propose mitigations.

Open Questions

• Which blockchain architectures provide the strongest non-repudiation guarantees?
• How do EU courts currently treat blockchain timestamps as evidence?
• What is the minimum viable on-chain record for insurance claim verification?

Insurance fraud involving manipulated or AI-generated imagery represents a rapidly growing category of financial crime. This theme studies the attack surface — how fraudulent claims are constructed using synthetic or edited images — and develops provenance-based countermeasures.

We work with anonymised claim datasets to characterise the statistical signatures of manipulated images in insurance contexts, and evaluate the marginal deterrent effect of mandatory provenance capture at the point of claim submission.

Open Questions

• What proportion of current fraudulent claims involve digitally manipulated imagery?
• What detection rate is achievable with current forensic methods?
• What legal and regulatory frameworks govern provenance requirements in insurance?

Detection methods for synthetic media age rapidly as generative models improve. This theme develops open, peer-reviewed forensic methodologies for detecting and attributing deepfakes — with an emphasis on techniques that remain valid across model generations and are robust to adversarial post-processing.

We focus on methodology rather than model benchmarking: the goal is a reproducible forensic science of synthetic media, not a leaderboard. We also study attribution — the provenance of generation, not just the fact of it.

Open Questions

• What forensic signatures survive adversarial post-processing?
• Can generation models be reliably attributed from their outputs?
• What evidentiary standard should forensic deepfake analysis meet in court?