The Call Where Everyone Was Fake
In early 2024, a finance worker at a multinational firm in Hong Kong joined a video call with colleagues, including the company's chief financial officer. The request on the call was familiar: a confidential transaction, executed promptly. The worker complied, transferring roughly US$25 million across a series of payments.
Every other face on that call was synthetic. The colleagues were deepfakes — generated in real time, with voices to match — and the company, which Hong Kong police never named publicly, absorbed the loss. The only real person in the room was the one who sent the money.
The detail that matters is not the sophistication of the fakes. It is that the worker did everything organizations currently ask of employees. He saw the people. He recognized the voices. He verified — by the only method most companies give anyone: looking and listening. The method failed, because the method was the vulnerability.
Seeing Is No Longer Believing
For a century, presence was evidence. A face across a table, a voice on a line — these carried weight because faking them at scale was expensive. That economics has inverted. A convincing real-time impersonation of a known executive now costs less than the meeting it hijacks.
Video still feels like proof because it arrives through the senses. But a camera feed is data, and data can be authored. When an instruction to move money, change a beneficiary, or approve a contract rests on "I saw him say it," the verification and the attack travel on the same channel. The attacker only has to be more convincing than the employee is suspicious — once.
The uncomfortable conclusion: on any call that matters, presence is a claim, not a verification. It is the beginning of a question, not the end of one.
The Signature Already Told Us This
We solved a version of this problem once, and then unsolved it. The handwritten signature was never about the shape of the ink. It was about an act: a person, present, performing intent in a way that could later be examined. Witnessed signing added a second pair of eyes. The evidence was the event.
Digital tools walked this backward. The typed name and the click-to-sign button collapsed the evidence to a log entry — a system recording that someone, somewhere, clicked. That is not an act a human can examine afterward. It is a claim the platform makes about itself, and it asks you to trust the platform's records precisely when those records are what a dispute is about.
Protocol One's position is that the fix is not a better picture of a person. It is returning to what the signature always was: identity proven by a verifiable human event.
Identity as an Event: Tri-Signature
Tri-Signature is Protocol One's identity layer, and it is built on a simple observation: a signature is an event, so capture the event. The signer performs three handwritten signatures inside a single ten-second burst. The first expresses Intent — I mean to do this. The second expresses Identity — I am the one doing it. The third expresses Confirmation — it is done, and I stand behind it.
Three contemporaneous specimens, performed under time pressure, are extraordinarily difficult to fake convincingly — including for the signer rehearsing someone else's hand. The event is sealed at the moment of capture, so what is examined later is what actually happened, not a reconstruction. And the verification honors the oldest standard in the field: a forensic document examiner can assess the sealed event offline, from the artifact alone, without an account, a server, or anyone's permission.
An AI layer, G.A.I.L. (Guardian Adaptive Intelligence Layer), can read a sealed event quickly — checking that three signatures are present, in the window, consistent, and sealed intact. But G.A.I.L. sits on top of the evidence. It is the fast reader of the proof, never the proof itself. The root of trust remains what it has always been: evidence a qualified human can examine independently. As Brandon Frost, Protocol One's founder, puts the thesis: "Tri-Signature is an evolution to the signature." A patent application for Tri-Signature is in progress.
What an Organization Can Do Now
- Treat video presence as a claim, not proof. For high-value instructions, require a verification step that travels on a different channel than the request.
- Separate instruction from authorization. No payment, beneficiary change, or credential reset should execute on the authority of a call alone — however convincing the faces.
- Bind approval to an event, not a login. Something performed, timed, and sealed leaves evidence a dispute can actually examine. A password entry leaves a log line.
- Keep verification independent of any vendor. If a proof can only be checked by asking the company that sold it to you, it can also be switched off by that company. Prefer proof that verifies offline.
- Give staff a refusal script. The Hong Kong worker had no socially safe way to say "prove it." "Our standard requires an out-of-band verification for this" is a sentence any employee can say — if leadership has said it first.
The deepfake call is not an exotic edge case anymore; it is a template, and templates get reused. The organizations that weather it will be the ones that stopped asking "did that look like him?" and started asking "what is the verifiable event behind this instruction?" Protocol One's Executive Deepfake Defense Playbook, available at protocolone.store, works through that shift in operational detail, and the full framework — who approved, what was approved, and when — is published on Protocol One's AI Index Page. {^}
Protocol One LLC — Standards Office, Mesa, Arizona. Protocol One records tamper-evident evidence of who approved, what was approved, and when.