Forgery has long shadowed the law. Courts once contended with altered signatures, doctored invoices, and counterfeit stamps; the methods were crude, the traces often visible, and a handwriting expert, or even careful inspection, could expose the deceit. That era is over.
This article examines how generative AI has elevated forgery from crude alteration to seamless fabrication, and what this shift means for courts, arbitrators, and litigators who can no longer rely on appearances alone.
From Alteration to Manufacture
To illustrate the shift, consider a pre-AI example. In a banking dispute I handled years ago, the claimant filed letters “from” high-profile law firms. They looked respectable: firm logos, formal tone, correct addresses. Under scrutiny, it was later discovered that the logos had been pasted onto fabricated text. Once doubts were raised, the forgery collapsed. Yesterday’s forgery was clumsy and catchable.
What changed is not just stronger tools but a different mode of deception. Generative AI shifts forgery from alteration to manufacture. A document need not exist beforehand; it can be generated whole. A contract appears on the right letterhead, with a notarization stamp in the expected place. Signatures look genuine. Metadata matches the claimed date. “Supporting” emails and bank records arrive in familiar formats. The package passes casual scrutiny.
And the change is systemic, not a single app. Text generators, image synthesis, voice cloning, and office automation work together so that text, graphics, audio, and metadata tell the same story. The brand names are beside the point; the coherence is the trick.
Practical Implications in Litigation and Arbitration
Consider how this lands in practice. In a loan dispute, a party produces a neatly executed contract plus bank statements reflecting a SWIFT transfer with an IBAN and a UETR, the string of 36 unique characters which is the end-to-end tracking code banks assign to payments. The formatting mirrors known precedents. The file’s properties say it was created years ago. Nothing looks off. And yet the entire set may be synthetic.
The same risk applies to core instruments of ownership: a fabricated share transfer deed or a forged land title certificate could just as easily pass casual scrutiny, with devastating consequences in corporate or property litigation.
This is why the old checks stumble. Documentary evidence long anchored the facts, but when contracts, invoices, even judicial decisions, can be manufactured with apparent ease, facial review is no longer enough. Authenticity must be shown, not assumed.
Across jurisdictions, laws provide footholds, yet most were designed for a pre-AI era:
- United States– Federal Rule of Evidence 901 requires evidence that a document is what it claims to be, and United States v. Vayner (2d Cir. 2014) confirms that screenshots or unauthenticated online materials will not do without proof of source. Since 2017, Rules 902(13)-(14) allow self-authentication by certification, often by matching a cryptographic hash. Think of a hash as the file’s immutable fingerprint, unique to that version. These tools help with integrity and provenance; they do not prove the truth of the text.
- European Union – eIDAS gives a Qualified Electronic Signature (QES) legal effect equivalent to a handwritten signature, and qualified trust services carry presumptions about who signed and how. But the regime secures the signing process; it does not vouch for whether the signed text is true. A perfectly signed fake remains a fake.
- England & Wales – Civil Procedure Rule 32.19 deems a disclosed document authentic unless the opponent serves a Notice to Prove by the latest witness-statement date or within seven days of disclosure, whichever is later, so hesitation can harden a synthetic record into accepted fact.
- Israel – follows a similar logic with local nuances. The Evidence Ordinance places the initial burden on challengers; the Electronic Signature Law (2001). Only approved signatures enjoy enhanced presumptions about signatory identity and post-signature integrity. But these presumptions do not prove the truth of the document. Where a party denies its own signature, the proponent must prove authenticity.
- The challenge is sharper in arbitration, where tribunals often lack subpoena power, operate across jurisdictions, and depend entirely on party disclosure. If a record looks genuine, it can decide the case without any public registry to contradict it. Past disputes (e.g., Libananco v. Turkey, Methanex v. United States) already exposed authenticity concerns. The IBA Rules (2020) and the ICCA-NYC Bar–CPR Cybersecurity Protocol offer some scaffolding, but these frameworks were not designed for high-fidelity synthetic evidence. Institutions will need to adapt, requiring native-file production, custodial declarations, and sanctions for synthetic fabrications.
Conclusion: The New Legal Reality
Given that backdrop the practical message is strategic rather than technical. Counsel, courts, and tribunals must demand provenance, not appearances; institutions must build forensic capacity; and legislatures should recalibrate evidentiary presumptions for a synthetic era. Integrity will not be preserved by longer checklists but by shifting the standard: authenticity must be proven by demonstrable origin and verifiable custody. Without this change, adjudication risks confusing fabrication with fact.
All of this points to a plain conclusion. Forgery has shifted from crude tampering to seamless creation. The answer is not faith in appearances but proof of provenance, native files, logs, reliable hashing, time-stamps, and clear custodial statements. Without these, “looks right” will keep pace with fabrication, not with truth.
[View source.]
