AI Forgery Risks
Generative tools can now copy handwriting, a face and a voice from a small sample. This guide explains the four risks that creates for a will, and what we rely on against each of them.
Four Risks, Described Concretely
The compliance page names four risks rather than talking about forgery in general. Each one is achievable with tools a member of the public can buy today.
- Fabricated handwritten wills: a generative model can reproduce handwriting from a sample. Where a handwritten-and-signed will is accepted without witnesses, a fabricated one can look authentic to a casual reader
- Deepfake identity video: a convincing video of a person can be produced with consumer tools. Without independent provenance, a video is content rather than evidence
- Voice-cloned instructions: a few minutes of recorded speech is enough to clone a voice, so any arrangement that depends on ringing someone to confirm an instruction stops being reliable
- Synthetic identity documents: generated passports and driving licences carry realistic security features and machine-readable zones, and an identity check without optical and cross-reference steps treats them as genuine
Why Paper and PDF Wills Are Exposed
What these risks have in common is that they attack documents with no record of how they came to exist. Three shapes are exposed in different ways.
A paper will found after death
There is no record of when or where it was written, on what paper, or with which pen or printer. A generated copy is hard to distinguish from the original because the original has no chain of evidence to compare it against.
A PDF signed with a pasted image
A file emailed round and signed by dropping in a picture of a signature carries no signing context: no device, no time, no identity check. Reproducing it needs nothing more than an image editor.
A video with nothing attached to it
A recording on its own proves only that somebody has the file. It is the timestamp, the originating address, the identity check and the audit trail around it that make a video worth anything.
What We Rely On, Risk by Risk
The approach is layered rather than clever: a successful forgery would have to defeat every layer at once, not just the most visible one.
- Against a fabricated handwritten will: every draft is captured with device-anchored provenance and an append-only, signed audit trail running from the first keystroke to the final signature. A handwritten copy has no equivalent record behind it
- Against a deepfake video: your government-issued identity document is machine-read and cross-checked against your profile, and the identity video is reviewed by a person rather than scored by a machine. A video that is not backed by that cross-check, a signed attestation and the audit trail does not complete the evidence pack
- Against a cloned voice: there is no path that accepts a will executed by voice, so a cloned voice has nothing to attack. Voice can support video evidence and never stands alone
- Against a synthetic identity document: the check reads the machine-readable zone, inspects optical security features and cross-references the extracted text, all on our own servers, and a mismatch flags the document. Your data never leaves our servers without your approval, so no external provider sees an identity document unless you specifically ask for enhanced processing of that document
No detector is perfect, and we do not claim one is
No deepfake or synthetic-document check catches everything, and saying otherwise would be marketing rather than safety. What layered evidence does is raise the cost: it increases how much a forgery has to fabricate before the pack holds together.
Three Things We Refuse To Do
These are deliberate limits rather than features not yet built. Each one would be convenient, and each would remove a layer.
- No execution by voice alone: a will is written and signed, and witnessed where that is required. A voice recording can supplement the evidence and can never replace it
- No third-party AI on your identity documents without your approval: verification runs on our own servers by default, and a document is only ever sent for enhanced external processing on your explicit approval for that document. You can decline and stay on our servers
- No silent fallback to a weaker path: if a verification service on our servers is unavailable, the flow pauses and asks you to try again later rather than quietly routing your document to a third party, so a failure is visible instead of hidden
This page describes our platform, not the law
Nothing here tells you whether a will is valid, what a court will accept as evidence, or what your jurisdiction requires of you. Rules on evidence and on execution differ from place to place and change. Speak to a qualified professional about your own situation.
Can’t find what you’re looking for?
Our support team is here to help. Contact us and we’ll get back to you as soon as possible.
Contact Support