By 2027, telling them apart unaided will not be possible.¹
Most viewers cannot tell the difference between authentic and synthetic media unaided — that is the state of the technology, not a personal failing. What follows is the operational record of the last twenty-four months. State, criminal, and commercial actors have moved from proof-of-concept to deployment, and the losses have been real.
Within hours of release, a synthetic video of a head of state appearing to call for surrender circulated across primary social platforms. Removed within six hours of detection — but viewed an estimated 2.1 million times in the interim.
A cloned voice impersonating a publicly-traded company's chief executive instructed treasury operations to move funds. Confirmed fabricated within ninety minutes; trades unwound, regulatory inquiry opened.
A synthetic robocall using the cloned voice of a sitting executive attempted to discourage voter turnout in a presidential primary. Cease-and-desist issued the same day; the precedent now anchors regulatory guidance across multiple jurisdictions.
In one week, video evidence in two unrelated criminal proceedings was challenged on the grounds that the footage may have been synthetic. One challenge was sustained, the other dismissed — the question of admissibility now sits unresolved on the public record.
Three documented cases across the MENA region in a single month: fabricated hostage videos used to extort ransom from families. Confirmed synthetic — but only after significant ransom transfers in two of the three. The emotional damage inflicted in the verification interval is, by definition, irreversible.
A regional state broadcaster aired mobile-phone footage of an alleged incident later determined to be synthetic. Correction issued within forty-eight hours. The footage had passed traditional human review. The broadcaster engaged with VERA's predecessor systems within two weeks.
The pattern is no longer emerging. It is operational. Adversaries — state, criminal, commercial — have moved from proof-of-concept to deployment. The institutions that will continue to be trusted are those that can prove what they broadcast, file, or act upon.
What you have seen so far is the threat. What follows is VERA's response. Seven detection layers unfold one at a time over a single specimen — and what they reveal is shown to you, not summarized for you.
A regional broadcaster has received a fourteen-second clip. The producer has no way to know whether what she is seeing is real. The bulletin airs in less than a minute. What follows is the next sixty seconds — in the language she would use to describe it to her editor, not the language an engineer would use to describe it to another engineer.
The producer can't watch every frame. She can't fact-check the source. She doesn't have time to find a second opinion. She sends the clip to VERA. What follows happens in the next 340 milliseconds.
VERA scans the frame for the statistical residue that synthetic generators always produce — patterns no human eye can see, but every generator leaves behind. It compares what it finds against a registry of known generators. In the first frame of this clip, VERA finds three fingerprints.
VERA reads how every part of the frame moves between one moment and the next. Eyes do not blink on schedule. Mouths do not move independently of jaws. Hair does not float. When motion violates physics, VERA flags it. In this clip, four flags cluster around the speaker's mouth.
Micro-tremor in the vowels. The small intake before a hard consonant. The slight rasp of an open throat. VERA listens for the texture that emerges from real human articulation — the texture current voice-synthesis still cannot reproduce. In this clip, none of it is there.
VERA searches every index it has access to — major platforms, archives, content-credential networks, watchlisted sources — and finds where this clip first surfaced. Fourteen minutes ago. On a forum the broadcaster has never published to. The clip has no credentials of authorship.
VERA puts the clip in front of Claude, Mistral, GPT, Gemini, and Google Vision, separately. Each returns its own verdict. When they disagree, VERA flags the clip for human review. When they agree, VERA has its answer. On this clip, all five return the same answer.
A clip was withheld. A bulletin went out. The broadcaster kept her audience's trust. This is what the discipline looks like — and what VERA is for.
What VERA did in those sixty seconds, it has done — with the same discipline and an auditable record — for every clip it has been asked to verify. The numbers follow.
What you just watched is one of 14,200 verifications VERA performed in the first quarter of 2026 alone. Across every one of them, the same discipline. Across every one of them, a record an editor — or a court — can audit.
What VERA did once, VERA does every time. The discipline is in the record.
VERA serves broadcasters who cannot afford to air the wrong clip, intelligence services who cannot afford to act on the wrong source, and forensic examiners who cannot afford to enter the wrong evidence. Same product. Three readings.
The lead bulletin airs at 14:33. A clip just landed in your producer's queue from a source she hasn't used before. She doesn't know if it's real.
VERA does — in 340 milliseconds, before she finishes reading the source line. Every step of the reasoning is on the record, so the producer can defend the call to her editor, the editor to the publisher, and the publisher to the board. Built around the pace of a real newsroom, not the pace of an engineering review.
The morning briefing is in ten minutes. The footage came in overnight, from an asset you've trusted for years. You cannot afford to act on a fabricated source. You also cannot afford to delay acting on a real one.
VERA returns a verdict your agency can defend to itself, in the time it takes to walk to the briefing room. Air-gapped where you need it. On your infrastructure, never ours. The full reasoning is preserved per specimen — auditable internally, citable in downstream products.
The case rests on a fourteen-second video. The defense is going to argue it could be fabricated. You need a verdict that holds up under cross-examination — and a methodology an expert witness can defend on the stand.
VERA is built for that day. Every verdict is reproducible — the same clip will always produce the same finding, with the same reasoning. The record is admissible. The methodology is defensible. Expert-witness support is available in jurisdictions where VERA's findings have been introduced.
VERA is in private preview with selected broadcast, intelligence, and forensic partners — by introduction or discreet inquiry. The next screen shows how to begin a conversation.
VERA is in private preview with a small number of broadcasters, intelligence services, and forensic operators — by introduction or discreet inquiry. We are not optimizing for volume of customers. We are optimizing for soundness of verdict. If you recognize yourself in any of what you have just read, the next step is here.
We respond personally. We do not respond at scale.