Synthetic Media, Deepfakes, and Digital Trust
An evidence-based account of what fails, what holds, and how each claim can be checked
Start here, by role
| Role | First action | Where the evidence is |
|---|---|---|
| Finance and treasury | Count the payment paths where one channel can both request and approve. The number is rarely zero and rarely known | Chapter 3, Chapter 12 |
| Information security | Rewrite “verify out of band” into three explicit properties: you initiate, to an address you already held, on a different channel | Chapter 3, Chapter 9 |
| Communications | Measure the rebuttal window: hours from a 02:00 fabricated statement to a live authoritative denial, counting approvals | Chapter 2, Chapter 12 |
| Legal and compliance | Establish whether originals of disputed recordings exist, and withdraw AI-text detection from decisions about people | Chapter 8, Chapter 11 |
| Board and audit | Stop accepting “no known evasions” as assurance. It is a guaranteed observation regardless of the truth | Chapter 7, Chapter 12 |
What to report instead of incident counts
A successful evasion produces no observable event, so a detector’s field failure rate cannot be measured from the defender’s side. Chapter 12 sets out six indicators computable from your own records without knowing whether you were attacked: the rebuttal window, the count of single-channel authorisation paths, provenance coverage at capture, original retrievability, red-team pass-through, and the number of decisions about people that touched a text detector.
What the evidence does not support
- Training staff to recognise synthetic media. Every verification heuristic consumed so far was a judgement about the artefact.
- Detection procurement for open-population content, where precision collapses and the errors are accusations against authentic material.
- Publishing detector verdicts, which is an attack surface in both directions.
- Voice biometrics as a primary authentication factor, given a three-second acquisition cost.
- Budgeting on legal deterrence of originators, given the collection record.
The full control set, ranked by the evidence behind each item with the claim that justifies it, is in Chapter 12. The cases those rankings derive from are collected under case studies.