AI content authentication: prove your work is human, before the model says otherwise
Generative AI surfaced two new evidentiary problems — proving a work is human, and proving it pre-dates the model accused of copying it. Blockchain timestamps under FRE 901(b)(9) address both, without an expert witness fee.
The two new evidentiary problems generative AI created
Three years ago, neither of these problems materially existed for working creators. Today, both show up in copyright disputes, platform moderation, and litigation.
Problem one — prove a piece of work is human. The U.S. Copyright Office issued guidance in March 2023 limiting copyright eligibility to works with human authorship; AI-generated content is not protected. Platforms have started enforcing related rules with automated AI-detection tools, and false positives against legitimately human content are increasingly reported.
Problem two — prove your work pre-dates the AI accused of copying it. Active litigation in this space — NYT v. OpenAI, the photographers' class action against Stability AI, the music labels against Suno and Udio — turns in part on whether plaintiffs' works existed before specific model training cutoffs. Forensic work to establish that timeline can take weeks of expert testimony.
What U.S. copyright law actually says about AI authorship
The U.S. Copyright Office's March 2023 guidance, refined further in subsequent reports, draws the line at human authorship. Where a human creator selects, arranges, or modifies AI-generated elements with sufficient creative judgment, the human-authored portions remain protectable. Purely machine-generated output is not. The key in disputes is the ability to show a clear, dated trail of human creative work.
This shifts the practical burden to creators. Where a manuscript, design, or codebase existed in full form before a generative model was capable of producing it — or before the model was even trained — that pre-existence is meaningful evidence. The cheaper and earlier you can establish it, the less expensive the dispute later.
How blockchain timestamps fit FRE 901(b)(9)
FRE 901(b)(9) authenticates evidence by "a process or system used to produce a result," with a showing that the process produces an accurate result. Blockchain anchoring lines up with the rule's structure: a SHA-256 hash sits in a block whose timestamp is established by the consensus of thousands of independent nodes, and any party can verify the proof against the public chain.
Combined with the ESIGN Act (15 U.S.C. § 7001) and state UETA frameworks, the path to admissibility on the date question is well-paved. The technical foundation needed at trial is a brief explanation of how the chain works and how the proof verifies — much shorter than the expert testimony required to establish pre-AI authorship by other means.
When to timestamp in a creator workflow
- Manuscripts and longer-form writing: timestamp the finished draft before submission to publishers, agents, or platforms that may run AI-detection.
- Visual design and illustration: timestamp final files plus key intermediate sketches; the sketch trail is itself strong evidence of human process.
- Source code: timestamp commits or releases independently of git's own timestamps (which a contributor controls).
- Photography: timestamp the RAW + metadata file before publication; RAW files are difficult to fabricate retroactively.
- Music composition: timestamp stems and DAW project files; pure renders are easier to mistake for AI output, project files less so.
How Bastamp works in practice
The SHA-256 hash is computed in your browser — the file itself never leaves your device (privacy by design). Bastamp aggregates the hash into a Merkle tree with other hashes from the same time window, then anchors the Merkle root on Polygon (primary anchor, low gas cost) and on Bitcoin via OpenTimestamps (secondary anchor, long-term integrity).
The output is a tamper-evident PDF certificate containing the document hash, the Merkle proof, the Polygon transaction hash, the Bitcoin block reference, and an explicit statement of the U.S. legal framework (FRE 901(b)(9), ESIGN Act, UETA). The certificate is independently verifiable against the blockchain forever — even if Bastamp ceases to exist.
Pricing starts at $2.99 per stamp, dropping to $0.94 per stamp in the 500-pack. No subscription, no smart card, no notary appointment. Your first stamp is free.
Legal framework — United States
Admissible under Federal Rule of Evidence 901(b)(9); self-authenticating under FRE 902(13) and 902(14). ESIGN Act and state UETA provide statutory recognition.
FAQ
Does a timestamp prove I wrote the file, or just that it existed?
It proves the file existed at a specific moment. Authorship is established by the surrounding context — drafts and revisions, contemporaneous metadata, your own testimony, and the absence of any earlier copy elsewhere. Timestamping is the strong backbone of an authorship case, not the entire case.
Can I prove pre-AI authorship of work I created two years ago?
Only forward, not backward. A timestamp attests that the file existed at the moment it was hashed, not before. If you have older work that wasn't timestamped, your evidence is whatever existed at the time (email attachments, cloud storage timestamps, server logs, drafts). Timestamping older work today proves it existed today, which is still useful against future AI-generation claims.
What if I used AI tools as part of my workflow but the final output is mine?
U.S. Copyright Office guidance protects human creative contributions even where AI tools are used. The practical burden is to show what you contributed. Timestamping intermediate human-edited drafts (the sketch trail, the prompt-and-revision history, the manual edits to AI-generated drafts) is much stronger evidence of human creative judgment than timestamping only the final output.
How is this different from copyright registration?
Copyright registration with the U.S. Copyright Office is a formal claim of authorship and a procedural prerequisite to certain damages in infringement litigation. It costs $35-65 and takes weeks to months. A blockchain timestamp is a contemporaneous evidence record. They complement each other: register what's commercially significant, timestamp the rest.
Does this work for code as well as text and images?
Yes. The hash is content-agnostic — any file becomes a SHA-256. For source code, timestamping a finished release file or zipped commit complements git's own timestamps (which any contributor can rewrite). For DMCA defense or claims that an AI model trained on your code, an externally-anchored timestamp is harder to challenge than a local git log.
What happens if a future AI can perfectly mimic my style?
The defense doesn't depend on style. It depends on dated existence: if your work was on-chain before the mimicking model existed or could have produced your output, the timestamp is dispositive on the date question. Style mimicry happens in copyright infringement disputes; timestamps move the dispute from a forensic battle to a documentary one.
Try your first stamp — free
No credit card. The certificate is admissible under the legal framework cited above.